
import os
import json
import pandas as pd
import numpy as np
import datetime
import warnings
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split, ShuffleSplit, cross_val_score, RandomizedSearchCV
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import roc_auc_score, accuracy_score, f1_score

%matplotlib inline

from sklearn.metrics import confusion_matrix, ConfusionMatrixDisplay
from sklearn.metrics import RocCurveDisplay
from sklearn.compose import ColumnTransformer
from sklearn.preprocessing import StandardScaler
from sklearn.preprocessing import OneHotEncoder
from sklearn.utils import resample


SEED = 10801
sns.set_style(style="whitegrid")
plt.rcParams["figure.figsize"] = 12, 8
warnings.filterwarnings("ignore")


PATH_TO_DATA = "../data"

df_train_features = pd.read_csv(os.path.join(PATH_TO_DATA, 
                                             "train_data.csv"), 
                                    index_col="match_id_hash")
df_train_targets = pd.read_csv(os.path.join(PATH_TO_DATA, 
                                            "train_targets.csv"), 
                                   index_col="match_id_hash")


df_train_features.shape

(31698, 245)


df_train_features.head()


df_train_targets.head()


X = df_train_features.values
y = df_train_targets["radiant_win"].values.astype("int8")


X_train, X_valid, y_train, y_valid = train_test_split(X, y, 
                                                      test_size=0.3, 
                                                      random_state=SEED)


%%time
rf_model = RandomForestClassifier(n_estimators=300, max_depth=7, n_jobs=-1, random_state=SEED)
rf_model.fit(X_train, y_train)

CPU times: user 30.6 s, sys: 62.5 ms, total: 30.6 s
Wall time: 4.86 s

RandomForestClassifier(max_depth=7, n_estimators=300, n_jobs=-1,
                       random_state=10801)


y_pred = rf_model.predict_proba(X_valid)[:, 1]


valid_score = roc_auc_score(y_valid, y_pred)
print("ROC-AUC score на отложенной части:", valid_score)

ROC-AUC score на отложенной части: 0.7754387258058622


valid_accuracy = accuracy_score(y_valid, y_pred > 0.5)
print("Accuracy score (p > 0.5) на отложенной части:", valid_accuracy)

Accuracy score (p > 0.5) на отложенной части: 0.6885383806519453


df_test_features = pd.read_csv(os.path.join(PATH_TO_DATA, "test_data.csv"), 
                                   index_col="match_id_hash")

X_test = df_test_features.values
y_test_pred = rf_model.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=df_test_features.index)


submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-09_18-13-59.csv


cv = ShuffleSplit(n_splits=5, test_size=0.3, random_state=SEED)


%%time
rf_model = RandomForestClassifier(n_estimators=300, max_depth=7, n_jobs=-1, random_state=SEED)
cv_scores_rf = cross_val_score(rf_model, X, y, cv=cv, scoring="roc_auc")

CPU times: user 8.36 s, sys: 675 ms, total: 9.04 s
Wall time: 31.8 s


cv_scores_rf

array([0.77543873, 0.77343884, 0.76649967, 0.7722681 , 0.77246001])


print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_rf.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.7720210676055513


with open(os.path.join(PATH_TO_DATA, "train_raw_data.jsonl")) as fin:
    # прочтем 419 строку
    for i in range(419):
        line = fin.readline()
    
    # переведем JSON в питоновский словарь 
    match = json.loads(line)


player = match["players"][9]
player["kills"], player["deaths"], player["assists"]

(0, 5, 5)


match.keys()

dict_keys(['game_time', 'match_id_hash', 'teamfights', 'objectives', 'chat', 'game_mode', 'lobby_type', 'players', 'targets'])


player.keys()

dict_keys(['player_slot', 'hero_id', 'hero_name', 'account_id_hash', 'ability_upgrades', 'obs_placed', 'sen_placed', 'creeps_stacked', 'camps_stacked', 'rune_pickups', 'firstblood_claimed', 'teamfight_participation', 'towers_killed', 'roshans_killed', 'observers_placed', 'stuns', 'max_hero_hit', 'times', 'gold_t', 'lh_t', 'dn_t', 'xp_t', 'obs_log', 'sen_log', 'obs_left_log', 'sen_left_log', 'purchase_log', 'kills_log', 'buyback_log', 'runes_log', 'obs', 'sen', 'actions', 'pings', 'purchase', 'gold_reasons', 'xp_reasons', 'killed', 'item_uses', 'ability_uses', 'hero_hits', 'damage', 'damage_taken', 'damage_inflictor', 'runes', 'killed_by', 'kill_streaks', 'multi_kills', 'life_state', 'healing', 'damage_inflictor_received', 'randomed', 'pred_vict', 'gold', 'lh', 'xp', 'x', 'y', 'hero_inventory', 'hero_stash', 'health', 'max_health', 'max_mana', 'level', 'kills', 'deaths', 'assists', 'denies', 'nearby_creep_death_count'])


player['nearby_creep_death_count']

164


sum(list(player['ability_uses'].values()))

62


sum(list(player["purchase"].values()))

36


for ab in player.keys():
    print(ab)

player_slot
hero_id
hero_name
account_id_hash
ability_upgrades
obs_placed
sen_placed
creeps_stacked
camps_stacked
rune_pickups
firstblood_claimed
teamfight_participation
towers_killed
roshans_killed
observers_placed
stuns
max_hero_hit
times
gold_t
lh_t
dn_t
xp_t
obs_log
sen_log
obs_left_log
sen_left_log
purchase_log
kills_log
buyback_log
runes_log
obs
sen
actions
pings
purchase
gold_reasons
xp_reasons
killed
item_uses
ability_uses
hero_hits
damage
damage_taken
damage_inflictor
runes
killed_by
kill_streaks
multi_kills
life_state
healing
damage_inflictor_received
randomed
pred_vict
gold
lh
xp
x
y
hero_inventory
hero_stash
health
max_health
max_mana
level
kills
deaths
assists
denies
nearby_creep_death_count


player["ability_uses"]

{'treant_living_armor': 52, 'treant_leech_seed': 5, 'treant_overgrowth': 5}


for i, player in enumerate(match["players"]):
    plt.plot(player["times"], player["xp_t"], label=str(i+1))

plt.legend()
plt.xlabel("Time, s")
plt.ylabel("XP")
plt.title("XP change for all players");


try:
    import ujson as json
except ModuleNotFoundError:
    import json
    print ("Подумайте об установке ujson, чтобы работать с JSON объектами быстрее")
    
try:
    from tqdm.notebook import tqdm
except ModuleNotFoundError:
    tqdm_notebook = lambda x: x
    print ("Подумайте об установке tqdm, чтобы следить за прогрессом")

    
def read_matches(matches_file, total_matches=31698, n_matches_to_read=None):
    """
    Аргуент
    -------
    matches_file: JSON файл с сырыми данными
    
    Результат
    ---------
    Возвращает записи о каждом матче
    """
    
    if n_matches_to_read is None:
        n_matches_to_read = total_matches
        
    c = 0
    with open(matches_file) as fin:
        for line in tqdm(fin, total=total_matches):
            if c >= n_matches_to_read:
                break
            else:
                c += 1
                yield json.loads(line)


def add_new_features(df_features, matches_file):
    """
    Аргуенты
    -------
    df_features: таблица с данными
    matches_file: JSON файл с сырыми данными
    
    Результат
    ---------
    Добавляет новые признаки в таблицу
    """
    
    for match in read_matches(matches_file):
        match_id_hash = match['match_id_hash']

        # Посчитаем количество разрушенных вышек обеими командами
        radiant_tower_kills = 0
        dire_tower_kills = 0
        for objective in match["objectives"]:
            if objective["type"] == "CHAT_MESSAGE_TOWER_KILL":
                if objective["team"] == 2:
                    radiant_tower_kills += 1
                if objective["team"] == 3:
                    dire_tower_kills += 1

        df_features.loc[match_id_hash, "radiant_tower_kills"] = radiant_tower_kills
        df_features.loc[match_id_hash, "dire_tower_kills"] = dire_tower_kills
        df_features.loc[match_id_hash, "diff_tower_kills"] = radiant_tower_kills - dire_tower_kills

        for player_id in range(10):
            player = match["players"][player_id]
            ability_upgr_n = len(player["ability_upgrades"])
            ability_uses_n = sum(list(player['ability_uses'].values()))
            purchase_n = sum(list(player["purchase"].values()))
            item_use_n = sum(list(player["item_uses"].values()))
            dmg = sum(list(player["damage"].values()))
            dmg_taken = sum(list(player["damage_taken"].values()))
            nearby_creep_death_count = player["nearby_creep_death_count"]
            pred_vict = player["pred_vict"]

            player_num = f"r{player_id + 1}" if player_id <= 4 else f"d{player_id - 4}"

            df_features.loc[match_id_hash, f"{player_num}_ability_upgrades"] = ability_upgr_n
            df_features.loc[match_id_hash, f"{player_num}_ability_uses"] = ability_uses_n
            df_features.loc[match_id_hash, f"{player_num}_purchases"] = purchase_n
            df_features.loc[match_id_hash, f"{player_num}_item_uses"] = item_use_n
            df_features.loc[match_id_hash, f"{player_num}_damage_given"] = dmg
            df_features.loc[match_id_hash, f"{player_num}_damage_taken"] = dmg_taken
            df_features.loc[match_id_hash, f"{player_num}_nearby_creep_death_count"] = nearby_creep_death_count
            df_features.loc[match_id_hash, f"{player_num}_pred_vict"] = pred_vict


            
            
        
        # ... (/¯◡ ‿ ◡)/¯☆*:・ﾟ добавляем новые признаки ...


# Скопируем таблицу с признаками
df_train_features_extended = df_train_features.copy()
df_test_features_extended = df_test_features.copy()

# Добавим новые
add_new_features(df_train_features_extended, 
                 os.path.join(PATH_TO_DATA, 
                              "train_raw_data.jsonl"))
add_new_features(df_test_features_extended, 
                 os.path.join(PATH_TO_DATA, 
                              "test_raw_data.jsonl"))

  0%|          | 0/31698 [00:00<?, ?it/s]

  0%|          | 0/31698 [00:00<?, ?it/s]


df_train_features_extended.to_csv("df_train_extended.tsv", sep="\t")
df_test_features_extended.to_csv("df_test_extended.tsv", sep="\t")


df_train_features_extended.head()


df_test_features_extended.head()


%%time
cv_scores_base = cross_val_score(rf_model, X, y, cv=cv, scoring="roc_auc", n_jobs=-1)
cv_scores_extended = cross_val_score(rf_model, df_train_features_extended.values, y, 
                                     cv=cv, scoring="roc_auc", n_jobs=-1)

CPU times: user 103 ms, sys: 228 ms, total: 331 ms
Wall time: 58.6 s


print(f"ROC-AUC на кросс-валидации для базовых признаков: {cv_scores_base.mean()}")
print(f"ROC-AUC на кросс-валидации для новых признаков: {cv_scores_extended.mean()}")

ROC-AUC на кросс-валидации для базовых признаков: 0.7720210676055513
ROC-AUC на кросс-валидации для новых признаков: 0.778629346381704


sns.histplot(df_train_features, x="game_time")
;

''


sns.heatmap(df_train_features.corr())
;

''


r1_ftrs_lst = [ftr for ftr in df_train_features.columns if ftr.startswith("r1")]
sns.heatmap(df_train_features[r1_ftrs_lst].corr() > 0.5, annot=True, fmt=".2f")

<AxesSubplot: >


list(df_train_features_extended.columns)

['game_time',
 'game_mode',
 'lobby_type',
 'objectives_len',
 'chat_len',
 'r1_hero_id',
 'r1_kills',
 'r1_deaths',
 'r1_assists',
 'r1_denies',
 'r1_gold',
 'r1_lh',
 'r1_xp',
 'r1_health',
 'r1_max_health',
 'r1_max_mana',
 'r1_level',
 'r1_x',
 'r1_y',
 'r1_stuns',
 'r1_creeps_stacked',
 'r1_camps_stacked',
 'r1_rune_pickups',
 'r1_firstblood_claimed',
 'r1_teamfight_participation',
 'r1_towers_killed',
 'r1_roshans_killed',
 'r1_obs_placed',
 'r1_sen_placed',
 'r2_hero_id',
 'r2_kills',
 'r2_deaths',
 'r2_assists',
 'r2_denies',
 'r2_gold',
 'r2_lh',
 'r2_xp',
 'r2_health',
 'r2_max_health',
 'r2_max_mana',
 'r2_level',
 'r2_x',
 'r2_y',
 'r2_stuns',
 'r2_creeps_stacked',
 'r2_camps_stacked',
 'r2_rune_pickups',
 'r2_firstblood_claimed',
 'r2_teamfight_participation',
 'r2_towers_killed',
 'r2_roshans_killed',
 'r2_obs_placed',
 'r2_sen_placed',
 'r3_hero_id',
 'r3_kills',
 'r3_deaths',
 'r3_assists',
 'r3_denies',
 'r3_gold',
 'r3_lh',
 'r3_xp',
 'r3_health',
 'r3_max_health',
 'r3_max_mana',
 'r3_level',
 'r3_x',
 'r3_y',
 'r3_stuns',
 'r3_creeps_stacked',
 'r3_camps_stacked',
 'r3_rune_pickups',
 'r3_firstblood_claimed',
 'r3_teamfight_participation',
 'r3_towers_killed',
 'r3_roshans_killed',
 'r3_obs_placed',
 'r3_sen_placed',
 'r4_hero_id',
 'r4_kills',
 'r4_deaths',
 'r4_assists',
 'r4_denies',
 'r4_gold',
 'r4_lh',
 'r4_xp',
 'r4_health',
 'r4_max_health',
 'r4_max_mana',
 'r4_level',
 'r4_x',
 'r4_y',
 'r4_stuns',
 'r4_creeps_stacked',
 'r4_camps_stacked',
 'r4_rune_pickups',
 'r4_firstblood_claimed',
 'r4_teamfight_participation',
 'r4_towers_killed',
 'r4_roshans_killed',
 'r4_obs_placed',
 'r4_sen_placed',
 'r5_hero_id',
 'r5_kills',
 'r5_deaths',
 'r5_assists',
 'r5_denies',
 'r5_gold',
 'r5_lh',
 'r5_xp',
 'r5_health',
 'r5_max_health',
 'r5_max_mana',
 'r5_level',
 'r5_x',
 'r5_y',
 'r5_stuns',
 'r5_creeps_stacked',
 'r5_camps_stacked',
 'r5_rune_pickups',
 'r5_firstblood_claimed',
 'r5_teamfight_participation',
 'r5_towers_killed',
 'r5_roshans_killed',
 'r5_obs_placed',
 'r5_sen_placed',
 'd1_hero_id',
 'd1_kills',
 'd1_deaths',
 'd1_assists',
 'd1_denies',
 'd1_gold',
 'd1_lh',
 'd1_xp',
 'd1_health',
 'd1_max_health',
 'd1_max_mana',
 'd1_level',
 'd1_x',
 'd1_y',
 'd1_stuns',
 'd1_creeps_stacked',
 'd1_camps_stacked',
 'd1_rune_pickups',
 'd1_firstblood_claimed',
 'd1_teamfight_participation',
 'd1_towers_killed',
 'd1_roshans_killed',
 'd1_obs_placed',
 'd1_sen_placed',
 'd2_hero_id',
 'd2_kills',
 'd2_deaths',
 'd2_assists',
 'd2_denies',
 'd2_gold',
 'd2_lh',
 'd2_xp',
 'd2_health',
 'd2_max_health',
 'd2_max_mana',
 'd2_level',
 'd2_x',
 'd2_y',
 'd2_stuns',
 'd2_creeps_stacked',
 'd2_camps_stacked',
 'd2_rune_pickups',
 'd2_firstblood_claimed',
 'd2_teamfight_participation',
 'd2_towers_killed',
 'd2_roshans_killed',
 'd2_obs_placed',
 'd2_sen_placed',
 'd3_hero_id',
 'd3_kills',
 'd3_deaths',
 'd3_assists',
 'd3_denies',
 'd3_gold',
 'd3_lh',
 'd3_xp',
 'd3_health',
 'd3_max_health',
 'd3_max_mana',
 'd3_level',
 'd3_x',
 'd3_y',
 'd3_stuns',
 'd3_creeps_stacked',
 'd3_camps_stacked',
 'd3_rune_pickups',
 'd3_firstblood_claimed',
 'd3_teamfight_participation',
 'd3_towers_killed',
 'd3_roshans_killed',
 'd3_obs_placed',
 'd3_sen_placed',
 'd4_hero_id',
 'd4_kills',
 'd4_deaths',
 'd4_assists',
 'd4_denies',
 'd4_gold',
 'd4_lh',
 'd4_xp',
 'd4_health',
 'd4_max_health',
 'd4_max_mana',
 'd4_level',
 'd4_x',
 'd4_y',
 'd4_stuns',
 'd4_creeps_stacked',
 'd4_camps_stacked',
 'd4_rune_pickups',
 'd4_firstblood_claimed',
 'd4_teamfight_participation',
 'd4_towers_killed',
 'd4_roshans_killed',
 'd4_obs_placed',
 'd4_sen_placed',
 'd5_hero_id',
 'd5_kills',
 'd5_deaths',
 'd5_assists',
 'd5_denies',
 'd5_gold',
 'd5_lh',
 'd5_xp',
 'd5_health',
 'd5_max_health',
 'd5_max_mana',
 'd5_level',
 'd5_x',
 'd5_y',
 'd5_stuns',
 'd5_creeps_stacked',
 'd5_camps_stacked',
 'd5_rune_pickups',
 'd5_firstblood_claimed',
 'd5_teamfight_participation',
 'd5_towers_killed',
 'd5_roshans_killed',
 'd5_obs_placed',
 'd5_sen_placed',
 'radiant_tower_kills',
 'dire_tower_kills',
 'diff_tower_kills',
 'd5_ability_upgrades',
 'd5_ability_uses',
 'd5_purchases',
 'd5_item_uses',
 'd5_damage_given',
 'd5_damage_taken']


for cmnd in ["r", "d"]:
    for plr in range(1, 6):
        plr_id = f"{cmnd}{plr}"
        plr_ftrs_lst = [ftr for ftr in df_train_features.columns if ftr.startswith(plr_id)]
        sns.heatmap(df_train_features[plr_ftrs_lst].corr() > 0.6, alpha=0.1, cbar=False)


sns.scatterplot(df_train_features[["d5_deaths", "d5_assists"]][1:100], x="d5_deaths", y="d5_assists")

<AxesSubplot: xlabel='d5_deaths', ylabel='d5_assists'>


PATH_TO_DATA = "../data"

df_train_features = pd.read_csv(os.path.join(PATH_TO_DATA, 
                                             "train_data.csv"), 
                                    index_col="match_id_hash")
df_train_targets = pd.read_csv(os.path.join(PATH_TO_DATA, 
                                            "train_targets.csv"), 
                                   index_col="match_id_hash")
df_test_features = pd.read_csv(os.path.join(PATH_TO_DATA, "test_data.csv"), 
                                   index_col="match_id_hash")


cv = ShuffleSplit(n_splits=5, test_size=0.3, random_state=SEED)


%%time 
# base model
X = df_train_features.values
y = df_train_targets["radiant_win"].values.astype("int8")
rf_model = RandomForestClassifier(n_estimators=300, max_depth=7, n_jobs=-1, random_state=SEED)
cv_scores_rf = cross_val_score(rf_model, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_rf.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.7720210676055513
CPU times: user 7.28 s, sys: 510 ms, total: 7.79 s
Wall time: 31.8 s


for team in ["r", "d"]:
    for player in range(1, 6):
        df_train_features[f"{team}{player}_kda"] = (df_train_features[f"{team}{player}_kills"] + df_train_features[f"{team}{player}_assists"]) / (df_train_features[f"{team}{player}_deaths"] + 1)
drop_list = []
for team in ["r", "d"]:
    for player in range(1, 6):
        drop_list.append(f"{team}{player}_kills")
        drop_list.append(f"{team}{player}_assists")
        drop_list.append(f"{team}{player}_deaths")
df_train_features = df_train_features.drop(drop_list, axis=1)

rad_first_blood_features_list = ["r1_firstblood_claimed", "r2_firstblood_claimed", "r3_firstblood_claimed", "r4_firstblood_claimed", "r5_firstblood_claimed"]
dire_first_blood_features_list = ["d1_firstblood_claimed", "d2_firstblood_claimed", "d3_firstblood_claimed", "d4_firstblood_claimed", "d5_firstblood_claimed"]
df_train_features["first_blood_claimed_radiant"] = df_train_features[rad_first_blood_features_list].sum(axis=1)
df_train_features = df_train_features.drop(rad_first_blood_features_list + dire_first_blood_features_list, axis=1)


for team in ["r", "d"]:
    for player in range(1, 6):
        df_test_features[f"{team}{player}_kda"] = (df_test_features[f"{team}{player}_kills"] + df_test_features[f"{team}{player}_assists"]) / (df_test_features[f"{team}{player}_deaths"] + 1)
drop_list = []
for team in ["r", "d"]:
    for player in range(1, 6):
        drop_list.append(f"{team}{player}_kills")
        drop_list.append(f"{team}{player}_assists")
        drop_list.append(f"{team}{player}_deaths")
df_test_features = df_test_features.drop(drop_list, axis=1)

rad_first_blood_features_list = ["r1_firstblood_claimed", "r2_firstblood_claimed", "r3_firstblood_claimed", "r4_firstblood_claimed", "r5_firstblood_claimed"]
dire_first_blood_features_list = ["d1_firstblood_claimed", "d2_firstblood_claimed", "d3_firstblood_claimed", "d4_firstblood_claimed", "d5_firstblood_claimed"]
df_test_features["first_blood_claimed_radiant"] = df_test_features[rad_first_blood_features_list].sum(axis=1)
df_test_features = df_test_features.drop(rad_first_blood_features_list + dire_first_blood_features_list, axis=1)


%%time
# kda and first blood claim by team
X = df_train_features.values
y = df_train_targets["radiant_win"].values.astype("int8")
rf_model = RandomForestClassifier(n_estimators=300, max_depth=7, n_jobs=-1, random_state=SEED)
cv_scores_rf = cross_val_score(rf_model, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_rf.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.7776936265657336
CPU times: user 7.66 s, sys: 561 ms, total: 8.22 s
Wall time: 25.3 s


for cmnd in ["r", "d"]:
    for plr in range(1, 6):
        plr_id = f"{cmnd}{plr}"
        plr_ftrs_lst = [ftr for ftr in df_train_features.columns if ftr.startswith(plr_id)]
        sns.heatmap(df_train_features[plr_ftrs_lst].corr() > 0.6, alpha=0.1, cbar=False)


r1_ftrs_lst = [ftr for ftr in df_train_features.columns if ftr.startswith("r1")]
sns.heatmap(df_train_features[r1_ftrs_lst].corr(), annot=True, fmt=".2f")

<AxesSubplot: >


rf_model = RandomForestClassifier(n_estimators=300, max_depth=7, n_jobs=-1, random_state=SEED)
rf_model.fit(X, y)
X_test = df_test_features.values
rf_model
y_test_pred = rf_model.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=df_test_features.index)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-05_23-17-55.csv


plr_ftrs_lst = [ftr[3:] for ftr in df_train_features.columns if ftr.startswith("r1")]
fig, axes = plt.subplots(7, 3, figsize=(15, 35))
axes = axes.ravel()
for feature, axis in enumerate(axes):
    r_feature = [f"r{plr}_{plr_ftrs_lst[feature]}" for plr in range(1, 6)]
    d_feature = [f"d{plr}_{plr_ftrs_lst[feature]}" for plr in range(1, 6)]
    sns.heatmap(df_train_features[r_feature + d_feature].corr(), ax=axis)


%%time
X = df_train_features.values
y = df_train_targets["radiant_win"].values.astype("int8")
rf_model = RandomForestClassifier(n_estimators=700, max_depth=19, n_jobs=-1, random_state=SEED)
cv_scores_rf = cross_val_score(rf_model, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_rf.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.787386574335213
CPU times: user 22.8 s, sys: 1.68 s, total: 24.5 s
Wall time: 2min 3s


rf_model = RandomForestClassifier(n_estimators=700, max_depth=19, n_jobs=-1, random_state=SEED)
rf_model.fit(X, y)
X_test = df_test_features.values
rf_model
y_test_pred = rf_model.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=df_test_features.index)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-06_00-11-19.csv


plrs = []
for tm in ["r", "d"]:
    for plr in range(1, 6):
        plrs.append(f"{tm}{plr}_hero_id")
plrs.append("radiant_win")

heroes_df = pd.concat([df_train_features, df_train_targets], axis=1)[plrs]

plrs.remove("radiant_win")

df_hero_count = pd.DataFrame(heroes_df.r1_hero_id.value_counts())\
    .reset_index()\
    .rename(columns={"index": "hero_id", "r1_hero_id": "games_total"})

for plr in plrs[1:]:
    intermed_df = pd.DataFrame(heroes_df[plr].value_counts())\
    .reset_index()\
    .rename(columns={"index": "hero_id", str(plr): "games_total"})
    df_hero_count = pd.merge(df_hero_count, intermed_df, on="hero_id")
df_hero_count["games_total"] = df_hero_count.drop("hero_id", axis=1).sum(axis=1)    
df_hero_count = df_hero_count[["hero_id", "games_total"]]
df_hero_count.head()


rad_win = heroes_df[heroes_df.radiant_win == True][['r1_hero_id', 'r2_hero_id', 'r3_hero_id', 'r4_hero_id', 'r5_hero_id']]
dire_win = heroes_df[heroes_df.radiant_win == False][['d1_hero_id', 'd2_hero_id', 'd3_hero_id', 'd4_hero_id', 'd5_hero_id']]
rad_win = rad_win.rename(columns={'r1_hero_id': "hero_1", 'r2_hero_id': "hero_2", 'r3_hero_id': "hero_3", 'r4_hero_id': "hero_4", 'r5_hero_id': "hero_5"})
dire_win = dire_win.rename(columns={'d1_hero_id': "hero_1", 'd2_hero_id': "hero_2", 'd3_hero_id': "hero_3", 'd4_hero_id': "hero_4", 'd5_hero_id': "hero_5"})
hero_win_df = pd.concat([rad_win, dire_win], axis=0)
hero_win_df.head()


df_win_count = pd.DataFrame(hero_win_df.hero_1.value_counts())\
    .reset_index()\
    .rename(columns={"index": "hero_id", "hero_1": "wins"})

for plr in range(2, 6):
    intermed_df = pd.DataFrame(hero_win_df[f"hero_{plr}"].value_counts())\
    .reset_index()\
    .rename(columns={"index": "hero_id", f"hero_{plr}": "wins"})
    df_win_count = pd.merge(df_win_count, intermed_df, on="hero_id")

df_win_count["win_count"] = df_win_count.drop("hero_id", axis=1).sum(axis=1)
df_win_count = df_win_count[["hero_id", "win_count"]]
df_win_count.head()


df_games_wins = pd.merge(df_hero_count, df_win_count, on="hero_id")
df_games_wins["win_prop"] = df_games_wins.win_count / df_games_wins.games_total
df_games_wins = df_games_wins.sort_values("hero_id")
win_proportion_dict = dict(zip(df_games_wins.hero_id, df_games_wins.win_prop))
df_games_wins.head()


for plr in plrs:
    df_train_features[plr] = df_train_features[plr].apply(lambda x: win_proportion_dict[x])
    df_test_features[plr] = df_test_features[plr].apply(lambda x: win_proportion_dict[x])


df_train_features.head()


df_test_features.head()


df_train_features


plrs_x = []
plrs_y = []
for tm in ["r", "d"]:
    for plr in range(1, 6):
        plrs_x.append(f"{tm}{plr}_x")
        plrs_y.append(f"{tm}{plr}_y")
x_min = min(df_train_features[plrs_x].min())
y_min = min(df_train_features[plrs_y].min())
x_max = max(df_train_features[plrs_x].max())
y_max = max(df_train_features[plrs_y].max())
print(f"Coordinates of the Ancient: ({x_min}, {y_min}); ({x_max}, {y_max}).")

Coordinates of the Ancient: (66, 70); (188, 186).


ancient_coord = (x_min, y_min)
plrs = []
for tm in ["r", "d"]:
    for plr in range(1, 6):
        plrs.append(f"{tm}{plr}")


def dists(x1, y1, x2, y2):
    return np.sqrt((np.square(x2 - x1) + np.square(y2 - y1)))

def distance_to_ancient(df, x2, y2):
    for plr in plrs:
        x1 = df[f"{plr}_x"]
        y1 = df[f"{plr}_y"]
        df[f"{plr}_dist_to_anc"] = dists(x1, y1, x2, y2)
    drop_lst = [f"{i}_x" for i in plrs] + [f"{i}_y" for i in plrs]
    df = df.drop(drop_lst, axis=1)
    return df


df_train_features = distance_to_ancient(df_train_features, ancient_coord[0], ancient_coord[1])
df_test_features = distance_to_ancient(df_test_features, ancient_coord[0], ancient_coord[1])


def mean_ftrs_by_plr(df, ftr):
    r_ftr_lst = [f"r{plr}_{ftr}" for plr in range(1, 6)]
    d_ftr_lst = [f"d{plr}_{ftr}" for plr in range(1, 6)]
    df[f'r_{ftr}'] = df[r_ftr_lst].mean(axis=1)
    df[f'd_{ftr}'] = df[d_ftr_lst].mean(axis=1)
    df = df.drop(r_ftr_lst + d_ftr_lst, axis=1)
    return (df)

def sum_up_ftrs_by_plr(df, ftr):
    r_ftr_lst = [f"r{plr}_{ftr}" for plr in range(1, 6)]
    d_ftr_lst = [f"d{plr}_{ftr}" for plr in range(1, 6)]
    df[f'r_{ftr}'] = df[r_ftr_lst].sum(axis=1)
    df[f'd_{ftr}'] = df[d_ftr_lst].sum(axis=1)
    df = df.drop(r_ftr_lst + d_ftr_lst, axis=1)
    return (df)


plr_ftrs_lst = [ftr[3:] for ftr in df_train_features.columns if ftr.startswith("r1")]

for feature in plr_ftrs_lst:
    df_train_features = sum_up_ftrs_by_plr(df=df_train_features, ftr=feature)
    df_test_features = sum_up_ftrs_by_plr(df=df_test_features, ftr=feature)


%%time
X = df_train_features.values
y = df_train_targets["radiant_win"].values.astype("int8")
rf_model = RandomForestClassifier(n_estimators=700, max_depth=19, n_jobs=-1, random_state=SEED)
cv_scores_rf = cross_val_score(rf_model, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_rf.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.819087237079156
CPU times: user 15.4 s, sys: 1.05 s, total: 16.5 s
Wall time: 1min 10s


df_train_features["r_gold_prop"] = df_train_features.r_gold / (df_train_features.r_gold + df_train_features.d_gold)
df_test_features["r_gold_prop"] = df_test_features.r_gold / (df_test_features.r_gold + df_test_features.d_gold)
df_train_features = df_train_features.drop(["r_gold", "d_gold"], axis=1)
df_test_features = df_test_features.drop(["r_gold", "d_gold"], axis=1)


sns.heatmap(df_train_features.corr() > 0.6)
;

''


per_time_norm_list = ['r_lh', 'd_lh',
       'r_xp', 'd_xp', 'r_health', 'd_health', 'r_max_health', 'd_max_health',
       'r_max_mana', 'd_max_mana', 'r_level', 'd_level', 'r_stuns', 'd_stuns', 
       'r_rune_pickups', 'd_rune_pickups', 'r_towers_killed', 'd_towers_killed',
       'r_obs_placed', 'd_obs_placed', 'r_sen_placed', 'd_sen_placed']


for ftr in per_time_norm_list:
    df_train_features[f"{ftr}_per_time"] = df_train_features[f"{ftr}"] / (df_train_features["game_time"] + 1)
    df_test_features[f"{ftr}_per_time"] = df_test_features[f"{ftr}"] / (df_test_features["game_time"] + 1)
    df_train_features = df_train_features.drop(f"{ftr}", axis=1)
    df_test_features = df_test_features.drop(f"{ftr}", axis=1)


sns.heatmap(df_train_features.drop(drop_list, axis=1).corr() > 0.8)
;

''


disbalance = ['lh_per_time',
 'xp_per_time',
 'health_per_time',
 'max_health_per_time',
 'max_mana_per_time',
 'level_per_time',
 'stuns_per_time',
 'rune_pickups_per_time',
 'obs_placed_per_time',
 'sen_placed_per_time']

for ftr in disbalance:
    df_train_features[f"r_{ftr}_prop"] = df_train_features[f"r_{ftr}"] / (df_train_features[f"r_{ftr}"] + df_train_features[f"d_{ftr}"] + 1)
    df_test_features[f"r_{ftr}_prop"] = df_test_features[f"r_{ftr}"] / (df_test_features[f"r_{ftr}"] + df_test_features[f"d_{ftr}"] + 1)
    df_train_features = df_train_features.drop([f"r_{ftr}", f"d_{ftr}"], axis=1)
    df_test_features = df_test_features.drop([f"r_{ftr}", f"d_{ftr}"], axis=1)


sns.heatmap(df_train_features.drop(drop_list, axis=1).corr() > 0.6)
;

''


sns.heatmap(df_train_features.corr())

<AxesSubplot: >


df_train_features.head()


df_train_features.columns

Index(['game_time', 'game_mode', 'lobby_type', 'objectives_len', 'chat_len',
       'first_blood_claimed_radiant', 'r_hero_id', 'd_hero_id', 'r_denies',
       'd_denies', 'r_creeps_stacked', 'd_creeps_stacked', 'r_camps_stacked',
       'd_camps_stacked', 'r_teamfight_participation',
       'd_teamfight_participation', 'r_roshans_killed', 'd_roshans_killed',
       'r_kda', 'd_kda', 'r_dist_to_anc', 'd_dist_to_anc', 'r_gold_prop',
       'r_towers_killed_per_time', 'd_towers_killed_per_time',
       'r_lh_per_time_prop', 'r_xp_per_time_prop', 'r_health_per_time_prop',
       'r_max_health_per_time_prop', 'r_max_mana_per_time_prop',
       'r_level_per_time_prop', 'r_stuns_per_time_prop',
       'r_rune_pickups_per_time_prop', 'r_obs_placed_per_time_prop',
       'r_sen_placed_per_time_prop'],
      dtype='object')


num_cols = df_train_features.columns.tolist()
num_cols.remove('first_blood_claimed_radiant')
cat_cols = ['first_blood_claimed_radiant']
preprocessor = ColumnTransformer(
    transformers=[("scaler", StandardScaler(), num_cols),
                  ("ohe", OneHotEncoder(drop="first"), cat_cols)]
)


df_train_targets.radiant_win.value_counts()

True     16670
False    15028
Name: radiant_win, dtype: int64


indexes = df_test_features.index


df_train = pd.concat([df_train_features, df_train_targets.drop("game_time", axis=1)], axis=1)
rad_win = df_train[df_train.radiant_win == True]
dir_win = df_train[df_train.radiant_win == False]
reshape_val = rad_win.shape[0]
dir_win_up = resample(dir_win, random_state=SEED, n_samples=reshape_val, replace=True)
df_train_balanced = pd.concat([rad_win, dir_win_up])


df_train_balanced.radiant_win.value_counts()

True     16670
False    16670
Name: radiant_win, dtype: int64


X, y = df_train_balanced.drop(['radiant_win', 'duration', 'time_remaining',
       'next_roshan_team'], axis=1), df_train_balanced["radiant_win"].values.astype("int8")


preprocessor.fit(X)
df_train_features_tr = preprocessor.transform(X)
df_test_features_tr = preprocessor.transform(df_test_features)


X.shape, len(y)

((33340, 35), 33340)


df_test_features.columns

Index(['game_time', 'game_mode', 'lobby_type', 'objectives_len', 'chat_len',
       'first_blood_claimed_radiant', 'r_hero_id', 'd_hero_id', 'r_denies',
       'd_denies', 'r_creeps_stacked', 'd_creeps_stacked', 'r_camps_stacked',
       'd_camps_stacked', 'r_teamfight_participation',
       'd_teamfight_participation', 'r_roshans_killed', 'd_roshans_killed',
       'r_kda', 'd_kda', 'r_dist_to_anc', 'd_dist_to_anc', 'r_gold_prop',
       'r_towers_killed_per_time', 'd_towers_killed_per_time',
       'r_lh_per_time_prop', 'r_xp_per_time_prop', 'r_health_per_time_prop',
       'r_max_health_per_time_prop', 'r_max_mana_per_time_prop',
       'r_level_per_time_prop', 'r_stuns_per_time_prop',
       'r_rune_pickups_per_time_prop', 'r_obs_placed_per_time_prop',
       'r_sen_placed_per_time_prop'],
      dtype='object')


%%time
# hero ids as win proportion and sumed up
X = df_train_features_tr
y = y
rf_model = RandomForestClassifier(n_estimators=700, max_depth=7, n_jobs=-1, random_state=SEED)
cv_scores_rf = cross_val_score(rf_model, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_rf.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8400976564665406
CPU times: user 8.05 s, sys: 764 ms, total: 8.81 s
Wall time: 39.9 s


# hero ids as win proportion and sumed up
rf_model = RandomForestClassifier(n_estimators=700, max_depth=7, n_jobs=-1, random_state=SEED)
rf_model.fit(X, y)
X_test = df_test_features_tr
rf_model
y_test_pred = rf_model.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=indexes)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-08_17-48-19.csv


ftr_imps = pd.DataFrame(list(zip(df_test_features.columns, rf_model.feature_importances_)), columns=["feature", "importance"])
sns.barplot(data=ftr_imps.sort_values("importance", ascending=False),
    x="importance", y="feature",
    linewidth=1.5, edgecolor="teal", facecolor='teal', alpha=0.5)
;

''


df_train_features_extended.head()


df_train_targets.head()


df_test_features_extended.head()


%%time 
# base model
X = df_train_features_extended.values
y = df_train_targets["radiant_win"].values.astype("int8")
rf_model = RandomForestClassifier(n_estimators=300, max_depth=7, n_jobs=-1, random_state=SEED)
cv_scores_rf = cross_val_score(rf_model, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_rf.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.7787220332559414
CPU times: user 22.6 s, sys: 1.45 s, total: 24 s
Wall time: 48.8 s


for team in ["r", "d"]:
    for player in range(1, 6):
        df_train_features_extended[f"{team}{player}_kda"] = (df_train_features_extended[f"{team}{player}_kills"] + df_train_features_extended[f"{team}{player}_assists"]) / (df_train_features_extended[f"{team}{player}_deaths"] + 1)
        df_test_features_extended[f"{team}{player}_kda"] = (df_test_features_extended[f"{team}{player}_kills"] + df_test_features_extended[f"{team}{player}_assists"]) / (df_test_features_extended[f"{team}{player}_deaths"] + 1)
drop_list = []
for team in ["r", "d"]:
    for player in range(1, 6):
        drop_list.append(f"{team}{player}_kills")
        drop_list.append(f"{team}{player}_assists")
        drop_list.append(f"{team}{player}_deaths")
df_train_features_extended = df_train_features_extended.drop(drop_list, axis=1)
df_test_features_extended = df_test_features_extended.drop(drop_list, axis=1)

rad_first_blood_features_list = ["r1_firstblood_claimed", "r2_firstblood_claimed", "r3_firstblood_claimed", "r4_firstblood_claimed", "r5_firstblood_claimed"]
dire_first_blood_features_list = ["d1_firstblood_claimed", "d2_firstblood_claimed", "d3_firstblood_claimed", "d4_firstblood_claimed", "d5_firstblood_claimed"]
df_train_features_extended["first_blood_claimed_radiant"] = df_train_features_extended[rad_first_blood_features_list].sum(axis=1)
df_train_features_extended = df_train_features_extended.drop(rad_first_blood_features_list + dire_first_blood_features_list, axis=1)
df_test_features_extended["first_blood_claimed_radiant"] = df_test_features_extended[rad_first_blood_features_list].sum(axis=1)
df_test_features_extended = df_test_features_extended.drop(rad_first_blood_features_list + dire_first_blood_features_list, axis=1)


%%time
# kda and first blood claim by team
X = df_train_features_extended.values
y = df_train_targets["radiant_win"].values.astype("int8")
rf_model = RandomForestClassifier(n_estimators=300, max_depth=7, n_jobs=-1, random_state=SEED)
cv_scores_rf = cross_val_score(rf_model, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_rf.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.7852818518586497
CPU times: user 8.13 s, sys: 524 ms, total: 8.66 s
Wall time: 38.3 s


for cmnd in ["r", "d"]:
    for plr in range(1, 6):
        plr_id = f"{cmnd}{plr}"
        plr_ftrs_lst = [ftr for ftr in df_train_features_extended.columns if ftr.startswith(plr_id)]
        sns.heatmap(df_train_features_extended[plr_ftrs_lst].corr() > 0.6, alpha=0.1, cbar=False)


df_train_features_extended


plrs = [f"{i}_hero_id" for i in plrs]
for plr in plrs:
    df_train_features_extended[plr] = df_train_features_extended[plr].apply(lambda x: win_proportion_dict[x])
    df_test_features_extended[plr] = df_test_features_extended[plr].apply(lambda x: win_proportion_dict[x])


ancient_coord = (x_min, y_min)
plrs = []
for tm in ["r", "d"]:
    for plr in range(1, 6):
        plrs.append(f"{tm}{plr}")


df_train_features_extended = distance_to_ancient(df_train_features_extended, ancient_coord[0], ancient_coord[1])
df_test_features_extended = distance_to_ancient(df_test_features_extended, ancient_coord[0], ancient_coord[1])


plr_ftrs_lst = [ftr[3:] for ftr in df_train_features_extended.columns if ftr.startswith("r1")]

for feature in plr_ftrs_lst:
    df_train_features_extended = sum_up_ftrs_by_plr(df=df_train_features_extended, ftr=feature)
    df_test_features_extended = sum_up_ftrs_by_plr(df=df_test_features_extended, ftr=feature)


%%time
# sumed featues
X = df_train_features_extended.values
y = df_train_targets["radiant_win"].values.astype("int8")
rf_model = RandomForestClassifier(n_estimators=700, max_depth=19, n_jobs=-1, random_state=SEED)
cv_scores_rf = cross_val_score(rf_model, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_rf.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8208474507971347
CPU times: user 16.1 s, sys: 979 ms, total: 17.1 s
Wall time: 1min 20s


sns.heatmap(df_train_features_extended.corr() > 0.6)
;

''


df_train_features_extended["r_gold_prop"] = df_train_features_extended.r_gold / (df_train_features_extended.r_gold + df_train_features_extended.d_gold + 1)
df_test_features_extended["r_gold_prop"] = df_test_features_extended.r_gold / (df_test_features_extended.r_gold + df_test_features_extended.d_gold + 1)
df_train_features_extended = df_train_features_extended.drop(["r_gold", "d_gold"], axis=1)
df_test_features_extended = df_test_features_extended.drop(["r_gold", "d_gold"], axis=1)


df_test_features_extended.columns

Index(['game_time', 'game_mode', 'lobby_type', 'objectives_len', 'chat_len',
       'radiant_tower_kills', 'dire_tower_kills', 'diff_tower_kills',
       'first_blood_claimed_radiant', 'r_hero_id', 'd_hero_id', 'r_denies',
       'd_denies', 'r_lh', 'd_lh', 'r_xp', 'd_xp', 'r_health', 'd_health',
       'r_max_health', 'd_max_health', 'r_max_mana', 'd_max_mana', 'r_level',
       'd_level', 'r_stuns', 'd_stuns', 'r_creeps_stacked', 'd_creeps_stacked',
       'r_camps_stacked', 'd_camps_stacked', 'r_rune_pickups',
       'd_rune_pickups', 'r_teamfight_participation',
       'd_teamfight_participation', 'r_towers_killed', 'd_towers_killed',
       'r_roshans_killed', 'd_roshans_killed', 'r_obs_placed', 'd_obs_placed',
       'r_sen_placed', 'd_sen_placed', 'r_ability_upgrades',
       'd_ability_upgrades', 'r_ability_uses', 'd_ability_uses', 'r_purchases',
       'd_purchases', 'r_item_uses', 'd_item_uses', 'r_damage_given',
       'd_damage_given', 'r_damage_taken', 'd_damage_taken',
       'r_nearby_creep_death_count', 'd_nearby_creep_death_count',
       'r_pred_vict', 'd_pred_vict', 'r_kda', 'd_kda', 'r_dist_to_anc',
       'd_dist_to_anc', 'r_gold_prop'],
      dtype='object')


per_time_norm_list = ['r_lh', 'd_lh',
       'r_xp', 'd_xp', 'r_health', 'd_health', 'r_max_health', 'd_max_health',
       'r_max_mana', 'd_max_mana', 'r_level', 'd_level', 'r_stuns', 'd_stuns', 
       'r_rune_pickups', 'd_rune_pickups', 'r_towers_killed', 'd_towers_killed',
       'r_obs_placed', 'd_obs_placed', 'r_sen_placed', 'd_sen_placed', 
       'r_ability_upgrades', 'd_ability_upgrades', 'r_ability_uses', 'd_ability_uses', 
       'r_purchases', 'd_purchases', 'r_item_uses', 'd_item_uses', 
       'r_damage_given', 'd_damage_given', 'r_damage_taken', 'd_damage_taken',
       'r_nearby_creep_death_count', 'd_nearby_creep_death_count']

for ftr in per_time_norm_list:
    df_train_features_extended[f"{ftr}_per_time"] = df_train_features_extended[f"{ftr}"] / (df_train_features_extended["game_time"] + 1)
    df_test_features_extended[f"{ftr}_per_time"] = df_test_features_extended[f"{ftr}"] / (df_test_features_extended["game_time"] + 1)
    df_train_features_extended = df_train_features_extended.drop(f"{ftr}", axis=1)
    df_test_features_extended = df_test_features_extended.drop(f"{ftr}", axis=1)


sns.heatmap(df_train_features_extended.corr() > 0.6)
;

''


disbalance = ['lh_per_time',
 'xp_per_time',
 'health_per_time',
 'max_health_per_time',
 'max_mana_per_time',
 'level_per_time',
 'stuns_per_time',
 'rune_pickups_per_time',
 'obs_placed_per_time',
 'sen_placed_per_time',
 'ability_upgrades_per_time',
 'ability_uses_per_time',
 'purchases_per_time',
 'item_uses_per_time',
 'damage_given_per_time',
 'damage_taken_per_time',
 'nearby_creep_death_count_per_time']

for ftr in disbalance:
    df_train_features_extended[f"r_{ftr}_prop"] = df_train_features_extended[f"r_{ftr}"] / (df_train_features_extended[f"r_{ftr}"] + df_train_features_extended[f"d_{ftr}"] + 1)
    df_test_features_extended[f"r_{ftr}_prop"] = df_test_features_extended[f"r_{ftr}"] / (df_test_features_extended[f"r_{ftr}"] + df_test_features_extended[f"d_{ftr}"] + 1)
    df_train_features_extended = df_train_features_extended.drop([f"r_{ftr}", f"d_{ftr}"], axis=1)
    df_test_features_extended = df_test_features_extended.drop([f"r_{ftr}", f"d_{ftr}"], axis=1)


sns.heatmap(df_train_features_extended.corr() > 0.8)
;

''


df_train_features_extended['pred_vict_difference'] = df_train_features_extended.r_pred_vict - df_train_features_extended.d_pred_vict
df_test_features_extended['pred_vict_difference'] = df_test_features_extended.r_pred_vict - df_test_features_extended.d_pred_vict
df_train_features_extended = df_train_features_extended.drop(['r_pred_vict', 'd_pred_vict'], axis=1)
df_test_features_extended = df_test_features_extended.drop(['r_pred_vict', 'd_pred_vict'], axis=1)


num_cols = df_train_features_extended.columns.tolist()
num_cols.remove('first_blood_claimed_radiant')
cat_cols = ['first_blood_claimed_radiant']
preprocessor = ColumnTransformer(
    transformers=[("scaler", StandardScaler(), num_cols),
                  ("ohe", OneHotEncoder(drop="first"), cat_cols)]
)

indexes = df_test_features.index

df_train_targets.radiant_win.value_counts()

True     16670
False    15028
Name: radiant_win, dtype: int64


df_train_ext = pd.concat([df_train_features_extended, df_train_targets.drop("game_time", axis=1)], axis=1)
rad_win = df_train_ext[df_train_ext.radiant_win == True]
dir_win = df_train_ext[df_train_ext.radiant_win == False]
reshape_val = rad_win.shape[0]
dir_win_up = resample(dir_win, random_state=SEED, n_samples=reshape_val, replace=True)
df_train_ext_balanced = pd.concat([rad_win, dir_win_up])
df_train_ext_balanced.radiant_win.value_counts()

True     16670
False    16670
Name: radiant_win, dtype: int64


X, y = df_train_ext_balanced.drop(['radiant_win', 'duration', 'time_remaining',
       'next_roshan_team'], axis=1), df_train_ext_balanced["radiant_win"].values.astype("int8")

preprocessor.fit(X)
df_train_features_tr = preprocessor.transform(X)
df_test_features_tr = preprocessor.transform(df_test_features_extended)

X.shape, len(y)

((33340, 46), 33340)


%%time
# extended dotaset
X = df_train_features_tr
y = y
rf_model = RandomForestClassifier(n_estimators=700, max_depth=7, n_jobs=-1, random_state=SEED)
cv_scores_rf = cross_val_score(rf_model, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_rf.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8408896301884274
CPU times: user 9.9 s, sys: 866 ms, total: 10.8 s
Wall time: 44 s


# extended dotaset BEST AS FAR on kaggle it gave 0.84015 score
rf_model = RandomForestClassifier(n_estimators=700, max_depth=7, n_jobs=-1, random_state=SEED)
rf_model.fit(X, y)
X_test = df_test_features_tr
rf_model
y_test_pred = rf_model.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=indexes)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-09_19-03-11.csv


ftr_imps = pd.DataFrame(list(zip(df_test_features_extended.columns, rf_model.feature_importances_)), columns=["feature", "importance"])
sns.barplot(data=ftr_imps.sort_values("importance", ascending=False),
    x="importance", y="feature",
    linewidth=1.5, edgecolor="teal", facecolor='teal', alpha=0.5)
;

''


import xgboost
import catboost


%%time
# extended dotaset
X = df_train_features_tr
y = y
cat_mdl = catboost.CatBoostClassifier(verbose=False, random_state=SEED,
                                  n_estimators=350, max_depth=11)
cv_scores_cb = cross_val_score(cat_mdl, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.912193335535936
CPU times: user 19min 10s, sys: 17.8 s, total: 19min 28s
Wall time: 3min 39s


%%time
# extended dotaset
X = df_train_features_tr
y = y
cat_mdl = catboost.CatBoostClassifier(verbose=False, random_state=SEED,
                                  n_estimators=700, max_depth=11)
cv_scores_cb = cross_val_score(cat_mdl, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.9150766864082801
CPU times: user 38min 54s, sys: 38.9 s, total: 39min 33s
Wall time: 7min 18s


# extended dotaset is worse than RF
cat_mdl = catboost.CatBoostClassifier(verbose=False, random_state=SEED,
                                  n_estimators=700, max_depth=11)
cat_mdl.fit(X, y)
X_test = df_test_features_tr
cat_mdl
y_test_pred = cat_mdl.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=indexes)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-09_20-16-30.csv


%%time
# extended dotaset
X = df_train_features_tr
y = y
xgb_mdl = xgboost.XGBRFClassifier(n_estimators=240, max_depth=7, random_state=SEED)
xgb_scores_cb = cross_val_score(xgb_mdl, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {xgb_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8478161514415096
CPU times: user 6min 31s, sys: 2.51 s, total: 6min 33s
Wall time: 53.7 s


%%time
# extended dotaset
X = df_train_features_tr
y = y
xgb_mdl = xgboost.XGBRFClassifier(n_estimators=700, max_depth=7, random_state=SEED)
xgb_scores_cb = cross_val_score(xgb_mdl, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {xgb_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8478865222757868
CPU times: user 21min 22s, sys: 11.7 s, total: 21min 34s
Wall time: 3min 2s


# extended dotaset slightly better than RF BEST AS FAR
xgb_mdl = xgboost.XGBRFClassifier(n_estimators=700, max_depth=7, random_state=SEED)
xgb_mdl.fit(X, y)
X_test = df_test_features_tr
xgb_mdl
y_test_pred = xgb_mdl.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=indexes)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-09_20-20-42.csv


rf = RandomForestClassifier()

rf_random = RandomizedSearchCV(estimator = rf, param_distributions = random_grid, cv = 3, verbose=2, random_state=42, n_jobs = -1)
# Fit the random search model
rf_random.fit(X_train, y_train)
rf_random.best_params_


random_grid = {
    'max_depth': [1, 5,  50, -1],
    'n_estimators':[1, 10 , 100],
    'learning_rate': np.linspace(0.01,0.5,4),
    'reg_alpha': [0, 0.2, 1],
    'reg_lambda': [0, 0.2, 1],
}

xgb_mdl = xgboost.XGBRFClassifier()

xgb_rand_search = RandomizedSearchCV(estimator = xgb_mdl, param_distributions = random_grid, cv = 3, verbose=2, random_state=42, n_jobs = -1)

xgb_rand_search.fit(X, y)
xgb_rand_search.best_params_

Fitting 3 folds for each of 10 candidates, totalling 30 fits
[CV] END learning_rate=0.5, max_depth=-1, n_estimators=100, reg_alpha=0, reg_lambda=0.2; total time=   0.1s
[CV] END learning_rate=0.5, max_depth=-1, n_estimators=100, reg_alpha=0, reg_lambda=0.2; total time=   0.1s
[CV] END learning_rate=0.5, max_depth=-1, n_estimators=100, reg_alpha=0, reg_lambda=0.2; total time=   0.1s
[CV] END learning_rate=0.01, max_depth=5, n_estimators=1, reg_alpha=0.2, reg_lambda=0; total time=   0.3s
[CV] END learning_rate=0.01, max_depth=5, n_estimators=1, reg_alpha=0.2, reg_lambda=0; total time=   0.3s
[CV] END learning_rate=0.01, max_depth=5, n_estimators=1, reg_alpha=0.2, reg_lambda=0; total time=   0.3s
[CV] END learning_rate=0.5, max_depth=50, n_estimators=10, reg_alpha=0.2, reg_lambda=1; total time=   5.2s
[CV] END learning_rate=0.5, max_depth=50, n_estimators=10, reg_alpha=0.2, reg_lambda=1; total time=   5.7s
[CV] END learning_rate=0.33666666666666667, max_depth=50, n_estimators=1, reg_alpha=0.2, reg_lambda=1; total time=   0.7s
[CV] END learning_rate=0.33666666666666667, max_depth=50, n_estimators=1, reg_alpha=0.2, reg_lambda=1; total time=   0.7s
[CV] END learning_rate=0.33666666666666667, max_depth=50, n_estimators=1, reg_alpha=0.2, reg_lambda=1; total time=   0.7s
[CV] END learning_rate=0.5, max_depth=-1, n_estimators=10, reg_alpha=0, reg_lambda=0; total time=   0.0s
[CV] END learning_rate=0.5, max_depth=-1, n_estimators=10, reg_alpha=0, reg_lambda=0; total time=   0.0s
[CV] END learning_rate=0.5, max_depth=-1, n_estimators=10, reg_alpha=0, reg_lambda=0; total time=   0.0s
[CV] END learning_rate=0.5, max_depth=50, n_estimators=10, reg_alpha=0.2, reg_lambda=1; total time=   5.4s
[CV] END learning_rate=0.17333333333333334, max_depth=5, n_estimators=100, reg_alpha=0, reg_lambda=0.2; total time=  14.1s
[CV] END learning_rate=0.17333333333333334, max_depth=5, n_estimators=100, reg_alpha=0, reg_lambda=0.2; total time=  14.9s
[CV] END learning_rate=0.17333333333333334, max_depth=-1, n_estimators=1, reg_alpha=1, reg_lambda=0; total time=   0.0s
[CV] END learning_rate=0.17333333333333334, max_depth=-1, n_estimators=1, reg_alpha=1, reg_lambda=0; total time=   0.0s
[CV] END learning_rate=0.17333333333333334, max_depth=-1, n_estimators=1, reg_alpha=1, reg_lambda=0; total time=   0.0s
[CV] END learning_rate=0.17333333333333334, max_depth=5, n_estimators=10, reg_alpha=0.2, reg_lambda=0.2; total time=   1.9s
[CV] END learning_rate=0.17333333333333334, max_depth=5, n_estimators=10, reg_alpha=0.2, reg_lambda=0.2; total time=   1.7s
[CV] END learning_rate=0.17333333333333334, max_depth=5, n_estimators=10, reg_alpha=0.2, reg_lambda=0.2; total time=   1.8s
[CV] END learning_rate=0.17333333333333334, max_depth=5, n_estimators=100, reg_alpha=0, reg_lambda=0.2; total time=  13.6s
[CV] END learning_rate=0.17333333333333334, max_depth=50, n_estimators=100, reg_alpha=0, reg_lambda=0; total time=  57.3s
[CV] END learning_rate=0.01, max_depth=50, n_estimators=100, reg_alpha=0.2, reg_lambda=0; total time=  57.9s
[CV] END learning_rate=0.01, max_depth=50, n_estimators=100, reg_alpha=0.2, reg_lambda=0; total time=  59.3s
[CV] END learning_rate=0.17333333333333334, max_depth=50, n_estimators=100, reg_alpha=0, reg_lambda=0; total time=  59.7s
[CV] END learning_rate=0.01, max_depth=50, n_estimators=100, reg_alpha=0.2, reg_lambda=0; total time=  59.2s
[CV] END learning_rate=0.17333333333333334, max_depth=50, n_estimators=100, reg_alpha=0, reg_lambda=0; total time=  59.4s

{'reg_lambda': 0,
 'reg_alpha': 0,
 'n_estimators': 100,
 'max_depth': 50,
 'learning_rate': 0.17333333333333334}


%%time
best_xgb = xgb_rand_search.best_estimator_
best_xgb_scores_cb = cross_val_score(best_xgb, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {best_xgb_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.9107475944056039
CPU times: user 7min 13s, sys: 3.84 s, total: 7min 17s
Wall time: 58.9 s


# extended dotaset + better params(?) slightly better than RF
xgb_mdl = best_xgb
xgb_mdl.fit(X, y)
X_test = df_test_features_tr
xgb_mdl
y_test_pred = xgb_mdl.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=indexes)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-09_20-44-23.csv


random_grid = {
    'max_depth': np.arange(1, 30, 5),
    'n_estimators': np.arange(50, 500, 10),
    'learning_rate': np.linspace(0.01,0.5,4),
    'reg_alpha': [0, 0.2, 1],
    'reg_lambda': [0, 0.2, 1],
}

xgb_mdl = xgboost.XGBRFClassifier()

xgb_rand_search = RandomizedSearchCV(estimator = xgb_mdl, param_distributions = random_grid, cv = 3, verbose=2, random_state=42, n_jobs = -1)

xgb_rand_search.fit(X, y)
xgb_rand_search.best_params_

Fitting 3 folds for each of 10 candidates, totalling 30 fits
[CV] END learning_rate=0.33666666666666667, max_depth=6, n_estimators=180, reg_alpha=1, reg_lambda=1; total time=  30.5s
[CV] END learning_rate=0.33666666666666667, max_depth=6, n_estimators=180, reg_alpha=1, reg_lambda=1; total time=  30.5s
[CV] END learning_rate=0.01, max_depth=11, n_estimators=100, reg_alpha=0.2, reg_lambda=1; total time=  34.8s
[CV] END learning_rate=0.01, max_depth=11, n_estimators=100, reg_alpha=0.2, reg_lambda=1; total time=  35.8s
[CV] END learning_rate=0.01, max_depth=11, n_estimators=100, reg_alpha=0.2, reg_lambda=1; total time=  38.2s
[CV] END learning_rate=0.33666666666666667, max_depth=1, n_estimators=410, reg_alpha=1, reg_lambda=0.2; total time=  13.8s
[CV] END learning_rate=0.33666666666666667, max_depth=1, n_estimators=410, reg_alpha=1, reg_lambda=0.2; total time=  11.9s
[CV] END learning_rate=0.33666666666666667, max_depth=1, n_estimators=410, reg_alpha=1, reg_lambda=0.2; total time=  13.1s
[CV] END learning_rate=0.33666666666666667, max_depth=6, n_estimators=180, reg_alpha=1, reg_lambda=1; total time=  35.6s
[CV] END learning_rate=0.33666666666666667, max_depth=11, n_estimators=120, reg_alpha=0, reg_lambda=0.2; total time=  48.7s
[CV] END learning_rate=0.33666666666666667, max_depth=11, n_estimators=120, reg_alpha=0, reg_lambda=0.2; total time=  47.9s
[CV] END learning_rate=0.33666666666666667, max_depth=11, n_estimators=120, reg_alpha=0, reg_lambda=0.2; total time=  48.3s
[CV] END learning_rate=0.01, max_depth=6, n_estimators=110, reg_alpha=1, reg_lambda=0.2; total time=  20.2s
[CV] END learning_rate=0.01, max_depth=6, n_estimators=110, reg_alpha=1, reg_lambda=0.2; total time=  21.1s
[CV] END learning_rate=0.01, max_depth=6, n_estimators=110, reg_alpha=1, reg_lambda=0.2; total time=  21.6s
[CV] END learning_rate=0.33666666666666667, max_depth=16, n_estimators=260, reg_alpha=0, reg_lambda=0.2; total time= 2.5min
[CV] END learning_rate=0.33666666666666667, max_depth=16, n_estimators=260, reg_alpha=0, reg_lambda=0.2; total time= 2.6min
[CV] END learning_rate=0.33666666666666667, max_depth=16, n_estimators=260, reg_alpha=0, reg_lambda=0.2; total time= 2.8min
[CV] END learning_rate=0.33666666666666667, max_depth=26, n_estimators=470, reg_alpha=1, reg_lambda=0.2; total time= 5.0min
[CV] END learning_rate=0.33666666666666667, max_depth=26, n_estimators=470, reg_alpha=1, reg_lambda=0.2; total time= 5.1min
[CV] END learning_rate=0.33666666666666667, max_depth=6, n_estimators=390, reg_alpha=1, reg_lambda=0.2; total time= 1.4min
[CV] END learning_rate=0.33666666666666667, max_depth=26, n_estimators=470, reg_alpha=1, reg_lambda=0.2; total time= 5.2min
[CV] END learning_rate=0.33666666666666667, max_depth=6, n_estimators=390, reg_alpha=1, reg_lambda=0.2; total time= 1.5min
[CV] END learning_rate=0.33666666666666667, max_depth=6, n_estimators=390, reg_alpha=1, reg_lambda=0.2; total time= 1.3min
[CV] END learning_rate=0.17333333333333334, max_depth=21, n_estimators=460, reg_alpha=1, reg_lambda=0.2; total time= 4.9min
[CV] END learning_rate=0.5, max_depth=11, n_estimators=290, reg_alpha=1, reg_lambda=0; total time= 1.8min
[CV] END learning_rate=0.5, max_depth=11, n_estimators=290, reg_alpha=1, reg_lambda=0; total time= 1.9min
[CV] END learning_rate=0.5, max_depth=11, n_estimators=290, reg_alpha=1, reg_lambda=0; total time= 1.8min
[CV] END learning_rate=0.17333333333333334, max_depth=21, n_estimators=460, reg_alpha=1, reg_lambda=0.2; total time= 4.8min
[CV] END learning_rate=0.17333333333333334, max_depth=21, n_estimators=460, reg_alpha=1, reg_lambda=0.2; total time= 4.3min

{'reg_lambda': 0.2,
 'reg_alpha': 0,
 'n_estimators': 260,
 'max_depth': 16,
 'learning_rate': 0.33666666666666667}


%%time
best_xgb = xgb_rand_search.best_estimator_
best_xgb_scores_cb = cross_val_score(best_xgb, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {best_xgb_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.9055695068283935
CPU times: user 19min 1s, sys: 8.88 s, total: 19min 9s
Wall time: 2min 38s


# extended dotaset + better params(?) slightly better than RF
xgb_mdl = best_xgb
xgb_mdl.fit(X, y)
X_test = df_test_features_tr
xgb_mdl
y_test_pred = xgb_mdl.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=indexes)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-09_21-01-37.csv


xgb_rand_search.best_estimator_

XGBRFClassifier(base_score=None, booster=None, callbacks=None,
                colsample_bylevel=None, colsample_bytree=None,
                early_stopping_rounds=None, enable_categorical=False,
                eval_metric=None, feature_types=None, gamma=None, gpu_id=None,
                grow_policy=None, importance_type=None,
                interaction_constraints=None, learning_rate=0.33666666666666667,
                max_bin=None, max_cat_threshold=None, max_cat_to_onehot=None,
                max_delta_step=None, max_depth=16, max_leaves=None,
                min_child_weight=None, missing=nan, monotone_constraints=None,
                n_estimators=260, n_jobs=None, num_parallel_tree=None,
                objective='binary:logistic', predictor=None, random_state=None, ...)


from hyperopt import fmin, hp, STATUS_OK, Trials, tpe


space = {'max_depth': hp.quniform('max_depth', 1, 18, 1),
        'gamma': hp.uniform ('gamma', 1, 9),
        'reg_alpha': hp.quniform('reg_alpha', 10, 100, 1),
        'reg_lambda': hp.uniform('reg_lambda', 0, 1),
        'colsample_bytree': hp.uniform('colsample_bytree', 0.5, 1),
        'min_child_weight': hp.quniform('min_child_weight', 0, 10, 1),
        'n_estimators': hp.uniform('n_estimators', 10, 100),
        'seed': SEED,
    }


X_tr, X_te, y_tr, y_te = train_test_split(X, y, test_size=0.25, random_state=SEED)


def objective(space):
    mdl = xgboost.XGBClassifier(n_estimators=int(space['n_estimators']), 
                              max_depth=int(space['max_depth']),
                              reg_alpha=int(space['reg_alpha']),
                              min_child_weight=int(space['min_child_weight']),
                              colsample_bytree=space['colsample_bytree'],
                              gamma=space['gamma'],
                              reg_lambda=space['reg_lambda'],
                             )
    
    evaluation = [(X_tr, y_tr), (X_te, y_te)]
    
    mdl.fit(X_tr, y_tr,
            eval_set=evaluation, eval_metric="auc",
            early_stopping_rounds=10,verbose=False)
    

    pred = mdl.predict(X_te)
    roc_auc = roc_auc_score(y_te, pred>0.5)
    print ("SCORE:", roc_auc)
    return {'loss': -roc_auc, 'status': STATUS_OK }


trials = Trials()

best_hyperparams = fmin(fn=objective,
                        space=space,
                        algo=tpe.suggest,
                        max_evals=300,
                        trials=trials)

SCORE:                                                 
0.7513634340883917                                     
SCORE:                                                                            
0.755151019723742                                                                 
SCORE:                                                                            
0.7483487634166751                                                               
SCORE:                                                                           
0.7518972046419697                                                               
SCORE:                                                                           
0.7559947912239934                                                               
SCORE:                                                                            
0.7703491565970344                                                                
SCORE:                                                                            
0.7580307721790461                                                                
SCORE:                                                                            
0.7601649618486065                                                                
SCORE:                                                                            
0.7578446046187981                                                                
SCORE:                                                                            
0.7582323145424448                                                                
SCORE:                                                                             
0.7648193811889341                                                                 
SCORE:                                                                             
0.7536932062257472                                                                 
SCORE:                                                                             
0.7518920797075863                                                                 
SCORE:                                                                             
0.7635502977183792                                                                 
SCORE:                                                                             
0.7594783358082735                                                                 
SCORE:                                                                             
0.7568155005195186                                                                 
SCORE:                                                                             
0.7574782293939057                                                                 
SCORE:                                                                             
0.7440342596681022                                                                 
SCORE:                                                                             
0.752429247689576                                                                  
SCORE:                                                                             
0.7588489190073796                                                                 
SCORE:                                                                             
0.7658441953150499                                                                 
SCORE:                                                                             
0.7654470704838559                                                                 
SCORE:                                                                             
0.7632718813392272                                                                 
SCORE:                                                                             
0.7598190863612483                                                                 
SCORE:                                                                             
0.7511465169218423                                                                 
SCORE:                                                                             
0.7706873446828172                                                                 
SCORE:                                                                             
0.7887524216754551                                                                 
SCORE:                                                                             
0.7893767135419654                                                                 
SCORE:                                                                             
0.7770352669192511                                                                 
SCORE:                                                                             
0.7994114502320501                                                                 
SCORE:                                                                             
0.7783120377913814                                                                 
SCORE:                                                                             
0.7755458688537353                                                                 
SCORE:                                                                             
0.8009520976414015                                                                 
SCORE:                                                                             
0.8012928481943763                                                                 
SCORE:                                                                             
0.756964958577862                                                                  
SCORE:                                                                             
0.7766381420880571                                                                 
SCORE:                                                                             
0.7701971360714994                                                                 
SCORE:                                                                             
0.7541219156244623                                                                 
SCORE:                                                                             
0.7697197397961404                                                                 
SCORE:                                                                             
0.763018254670773                                                                  
SCORE:                                                                             
0.7683985720205301                                                                 
SCORE:                                                                             
0.7775562251368703                                                                 
SCORE:                                                                             
0.7681278430429535                                                                 
SCORE:                                                                             
0.795982552420017                                                                  
SCORE:                                                                             
0.7621181088923025                                                                 
SCORE:                                                                             
0.7718718667360432                                                                 
SCORE:                                                                             
0.759576544522781                                                                  
SCORE:                                                                             
0.7511516418562258                                                                 
SCORE:                                                                             
0.7615220041649018                                                                 
SCORE:                                                                             
0.76495942433973                                                                   
SCORE:                                                                             
0.75321837241758                                                                   
SCORE:                                                                             
0.7749062021840742                                                                 
SCORE:                                                                             
0.7546736234483828                                                                 
SCORE:                                                                             
0.757832627244059                                                                  
SCORE:                                                                             
0.7779379751649135                                                                 
SCORE:                                                                             
0.767748655482102                                                                  
SCORE:                                                                             
0.7809108114003417                                                                 
SCORE:                                                                             
0.7570606048251776                                                                 
SCORE:                                                                             
0.756484999835311                                                                  
SCORE:                                                                             
0.7709452613244351                                                                 
SCORE:                                                                             
0.7670543420401937                                                                 
SCORE:                                                                             
0.7563611664488857                                                                 
SCORE:                                                                             
0.7524753720990279                                                                 
SCORE:                                                                             
0.7532465595566897                                                                 
SCORE:                                                                             
0.7679160508107876                                                                 
SCORE:                                                                             
0.7969527197738857                                                                 
SCORE:                                                                             
0.7956759489017553                                                                 
SCORE:                                                                             
0.7868300242380606                                                                 
SCORE:                                                                             
0.7885073173697962                                                                 
SCORE:                                                                             
0.7935648214369211                                                                 
SCORE:                                                                             
0.767807592227513                                                                  
SCORE:                                                                             
0.8063682895318436                                                                 
SCORE:                                                                             
0.8172339853862208                                                                 
SCORE:                                                                             
0.7773316205687455                                                                 
SCORE:                                                                             
0.8057721848044428                                                                 
SCORE:                                                                             
0.8134054002758021                                                                 
SCORE:                                                                             
0.7834839603373842                                                                 
SCORE:                                                                             
0.7941660510987054                                                                 
SCORE:                                                                             
0.7695702817377972                                                                 
SCORE:                                                                             
0.7778892882882696                                                                 
SCORE:                                                                             
0.752836622389537                                                                  
SCORE:                                                                             
0.8117793564879021                                                                 
SCORE:                                                                             
0.7530689143592368                                                                 
SCORE:                                                                             
0.7845207518382391                                                                 
SCORE:                                                                             
0.753419079819759                                                                  
SCORE:                                                                             
0.8036046830613893                                                                 
SCORE:                                                                             
0.7650337358882916                                                                 
SCORE:                                                                             
0.8007975146486748                                                                 
SCORE:                                                                             
0.7538093522105973                                                                 
SCORE:                                                                             
0.7577651681358528                                                                 
SCORE:                                                                             
0.7787561219932471                                                                 
SCORE:                                                                             
0.7824360840065423                                                                 
SCORE:                                                                             
0.76310108858213                                                                   
SCORE:                                                                             
0.8045697254808744                                                                 
SCORE:                                                                             
0.7512601004395005                                                                 
SCORE:                                                                             
0.7512806001770348                                                                 
SCORE:                                                                             
0.7589522526562708                                                                 
SCORE:                                                                             
0.7891264843019228                                                                 
SCORE:                                                                             
0.7597388149170831                                                                 
SCORE:                                                                             
0.7556617280725939                                                                 
SCORE:                                                                              
0.7979203246605624                                                                  
SCORE:                                                                              
0.7599711068867832                                                                  
SCORE:                                                                              
0.7699674065689912                                                                  
SCORE:                                                                              
0.7554584582032232                                                                  
SCORE:                                                                              
0.7574372299188371                                                                  
SCORE:                                                                              
0.7612615250560921                                                                  
SCORE:                                                                              
0.7531978726800458                                                                  
SCORE:                                                                              
0.7757089742092572                                                                  
SCORE:                                                                              
0.7939747586040738                                                                  
SCORE:                                                                              
0.7636946308423389                                                                  
SCORE:                                                                              
0.7980108459734945                                                                  
SCORE:                                                                              
0.7580000225727447                                                                  
SCORE:                                                                              
0.7735927218100394                                                                  
SCORE:                                                                              
0.7672686967395512                                                                  
SCORE:                                                                              
0.7734039917825996                                                                  
SCORE:                                                                              
0.7593750021593824                                                                  
SCORE:                                                                              
0.794493989315721                                                                   
SCORE:                                                                              
0.7593186278811632                                                                  
SCORE:                                                                              
0.7559794164208424                                                                  
SCORE:                                                                              
0.7868556489099784                                                                  
SCORE:                                                                              
0.8095401056634786                                                                  
SCORE:                                                                              
0.8038241626951304                                                                  
SCORE:                                                                              
0.7805495611098329                                                                  
SCORE:                                                                              
0.7803172691401328                                                                  
SCORE:                                                                              
0.7539126858594883                                                                  
SCORE:                                                                              
0.7693021152274123                                                                  
SCORE:                                                                              
0.7843038346716898                                                                  
SCORE:                                                                              
0.7915519026865943                                                                  
SCORE:                                                                              
0.7686103642526959                                                                  
SCORE:                                                                              
0.7879240249783543                                                                  
SCORE:                                                                              
0.7518510802325177                                                                  
SCORE:                                                                              
0.8187336333205036                                                                  
SCORE:                                                                              
0.767410467396319                                                                   
SCORE:                                                                              
0.7559922287568015                                                                  
SCORE:                                                                              
0.7872066493317201                                                                  
SCORE:                                                                              
0.7620250251121785                                                                  
SCORE:                                                                              
0.7612538376545168                                                                  
SCORE:                                                                              
0.7674394894966485                                                                  
SCORE:                                                                              
0.753953685334557                                                                   
SCORE:                                                                              
0.7770275795176756                                                                  
SCORE:                                                                              
0.7523566636469862                                                                  
SCORE:                                                                              
0.7541629150995307                                                                  
SCORE:                                                                              
0.7711954905644776                                                                  
SCORE:                                                                              
0.7839015849061124                                                                  
SCORE:                                                                              
0.7577130838307975                                                                  
SCORE:                                                                              
0.8080917070730314                                                                  
SCORE:                                                                              
0.81173323207845                                                                    
SCORE:                                                                              
0.7908242771711927                                                                  
SCORE:                                                                              
0.8203178426721155                                                                  
SCORE:                                                                              
0.8049779351420551                                                                  
SCORE:                                                                              
0.8034757247405804                                                                  
SCORE:                                                                              
0.8092762291262575                                                                  
SCORE:                                                                              
0.8058960181908683                                                                  
SCORE:                                                                              
0.8158461934636241                                                                  
SCORE:                                                                              
0.8042417872638585                                                                  
SCORE:                                                                              
0.8080327703276204                                                                  
SCORE:                                                                              
0.8062880180876786                                                                  
SCORE:                                                                              
0.7588489190073796                                                                  
SCORE:                                                                              
0.7828537085752706                                                                  
SCORE:                                                                              
0.8251456172367321                                                                  
SCORE:                                                                              
0.7820201869437863                                                                  
SCORE:                                                                              
0.8070626029737518                                                                  
SCORE:                                                                              
0.7774144544801026                                                                  
SCORE:                                                                              
0.7781318301267368                                                                  
SCORE:                                                                              
0.7968288863874602                                                                  
SCORE:                                                                              
0.7976752203549035                                                                  
SCORE:                                                                              
0.765134507069991                                                                   
SCORE:                                                                              
0.7808800617940402                                                                  
SCORE:                                                                              
0.7871758997254188                                                                  
SCORE:                                                                              
0.7567616887084911                                                                  
SCORE:                                                                              
0.7698384482481823                                                                  
SCORE:                                                                              
0.8061411224965271                                                                  
SCORE:                                                                              
0.7788543307077547                                                                  
SCORE:                                                                              
0.7540493315818726                                                                  
SCORE:                                                                              
0.7946613846444066                                                                  
SCORE:                                                                              
0.7793189146471546                                                                  
SCORE:                                                                              
0.7552210412991398                                                                  
SCORE:                                                                              
0.761542503902436                                                                   
SCORE:                                                                              
0.7883988587865214                                                                  
SCORE:                                                                              
0.7750300355704995                                                                  
SCORE:                                                                              
0.7861519205605226                                                                  
SCORE:                                                                              
0.7823711873655281                                                                  
SCORE:                                                                              
0.812208065886617                                                                   
SCORE:                                                                              
0.7961969071193746                                                                  
SCORE:                                                                              
0.8094675216208888                                                                  
SCORE:                                                                              
0.7768311620886607                                                                  
SCORE:                                                                              
0.7548828532133569                                                                  
SCORE:                                                                              
0.7995796805219555                                                                  
SCORE:                                                                              
0.7656349655500759                                                                  
SCORE:                                                                              
0.7654778200901574                                                                  
SCORE:                                                                              
0.7877976291247373                                                                  
SCORE:                                                                              
0.7671576756890848                                                                  
SCORE:                                                                              
0.779569143887197                                                                   
SCORE:                                                                              
0.7527153514703034                                                                  
SCORE:                                                                              
0.8108963130185541                                                                  
SCORE:                                                                              
0.788532942041714                                                                   
SCORE:                                                                              
0.7692346561192062                                                                  
SCORE:                                                                              
0.7724782213322111                                                                  
SCORE:                                                                              
0.7650106736835656                                                                  
SCORE:                                                                              
0.7847479188735555                                                                  
SCORE:                                                                              
0.7613751085737505                                                                  
SCORE:                                                                              
0.791740632714034                                                                   
SCORE:                                                                              
0.8059455400287319                                                                  
SCORE:                                                                              
0.762252192147495                                                                   
SCORE:                                                                              
0.7782505385787785                                                                  
SCORE:                                                                              
0.7616304627481764                                                                  
SCORE:                                                                              
0.7938304254801142                                                                  
SCORE:                                                                              
0.7590965857802303                                                                  
SCORE:                                                                              
0.7885406294432893                                                                  
SCORE:                                                                              
0.7578890015222783                                                                  
SCORE:                                                                              
0.8159956515219673                                                                  
SCORE:                                                                              
0.770935011455668                                                                   
SCORE:                                                                              
0.8038292876295139                                                                  
SCORE:                                                                              
0.7711801157613268                                                                  
SCORE:                                                                              
0.7596978154420146                                                                  
SCORE:                                                                              
0.7541577901651473                                                                  
SCORE:                                                                              
0.7665436336913417                                                                  
SCORE:                                                                              
0.7553499996199486                                                                  
SCORE:                                                                              
0.804933538238575                                                                   
SCORE:                                                                              
0.7750479728408421                                                                  
SCORE:                                                                              
0.7935391967650032                                                                  
SCORE:                                                                              
0.7623888378698792                                                                  
SCORE:                                                                              
0.7807587908748067                                                                  
SCORE:                                                                              
0.7586832511846661                                                                  
SCORE:                                                                              
0.8116913976421616                                                                  
SCORE:                                                                              
0.7553986864965925                                                                  
SCORE:                                                                              
0.7708478875711473                                                                  
SCORE:                                                                              
0.8005677851461667                                                                  
SCORE:                                                                              
0.7552492284382494                                                                  
SCORE:                                                                              
0.7711749908269433                                                                  
SCORE:                                                                              
0.7812848740268095                                                                  
SCORE:                                                                              
0.7708991369149829                                                                  
SCORE:                                                                              
0.8081899157875388                                                                  
SCORE:                                                                              
0.7528417473239205                                                                  
SCORE:                                                                              
0.7683857596845712                                                                  
SCORE:                                                                              
0.810664856010074                                                                   
SCORE:                                                                              
0.7579795228352104                                                                  
SCORE:                                                                              
0.795286511472137                                                                   
SCORE:                                                                              
0.7958348218676459                                                                  
SCORE:                                                                              
0.7618806919882188                                                                  
SCORE:                                                                              
0.8125744411115096                                                                  
SCORE:                                                                              
0.7551228325846323                                                                  
SCORE:                                                                              
0.7687444475078883                                                                  
SCORE:                                                                              
0.7567950007819844                                                                  
SCORE:                                                                              
0.7582656266159378                                                                  
SCORE:                                                                              
0.757914626194196                                                                   
SCORE:                                                                              
0.7695907814753316                                                                  
SCORE:                                                                              
0.761971213301151                                                                   
SCORE:                                                                              
0.7540595814506398                                                                  
SCORE:                                                                              
0.7669100089162342                                                                  
SCORE:                                                                              
0.7769165584672093                                                                  
SCORE:                                                                              
0.8020486608488873                                                                  
SCORE:                                                                              
0.7570554798907939                                                                  
SCORE:                                                                              
0.8086673120628979                                                                  
SCORE:                                                                              
0.795017510000532                                                                   
SCORE:                                                                              
0.7899369437286812                                                                  
SCORE:                                                                              
0.7611479415384339                                                                  
SCORE:                                                                              
0.8154021092617584                                                                  
SCORE:                                                                              
0.815219339129922                                                                   
SCORE:                                                                              
0.8018274537091741                                                                  
SCORE:                                                                              
0.8082906869692381                                                                  
SCORE:                                                                              
0.8117793564879021                                                                  
SCORE:                                                                              
0.813861461852407                                                                   
SCORE:                                                                              
0.8000673266660816                                                                  
SCORE:                                                                              
0.8107912518636909                                                                  
SCORE:                                                                              
0.7946921342507081                                                                  
SCORE:                                                                              
0.8011818271439098                                                                  
SCORE:                                                                              
0.7993730132241734                                                                  
SCORE:                                                                              
0.7921505698811868                                                                  
SCORE:                                                                              
0.7944675296825835                                                                  
SCORE:                                                                              
0.790124838794901                                                                   
SCORE:                                                                              
0.802877057545988                                                                   
SCORE:                                                                              
0.8090183124846396                                                                  
SCORE:                                                                              
0.7987999707014987                                                                  
SCORE:                                                                              
0.802340724525218                                                                   
SCORE:                                                                              
0.7911086534459483                                                                  
SCORE:                                                                              
0.7886285882890297                                                                  
SCORE:                                                                              
0.8065074977214195                                                                  
SCORE:                                                                              
0.810210521939441                                                                   
SCORE:                                                                              
0.7834959377121232                                                                  
SCORE:                                                                              
0.8050061222811647                                                                  
SCORE:                                                                              
0.8076433328980018                                                                  
SCORE:                                                                              
0.7871084406172126                                                                  
SCORE:                                                                              
0.8005549728102076                                                                  
SCORE:                                                                              
0.8023586617955605                                                                  
SCORE:                                                                              
0.7883399220411105                                                                  
SCORE:                                                                              
0.8143960673672048                                                                  
SCORE:                                                                              
0.7582861263534721                                                                  
SCORE:                                                                              
0.7806554572259157                                                                  
SCORE:                                                                              
0.7865874823995933                                                                  
SCORE:                                                                              
0.7835864590250556                                                                  
SCORE:                                                                              
0.8070113536299162                                                                  
SCORE:                                                                              
0.7557394370495671                                                                  
SCORE:                                                                              
0.8117844814222857                                                                  
SCORE:                                                                              
0.8035559961847455                                                                  
SCORE:                                                                              
0.7787663718620141                                                                  
SCORE:                                                                              
0.7808228525546013                                                                  
SCORE:                                                                              
0.7774298292832532                                                                  
SCORE:                                                                              
0.7969552822410775                                                                  
SCORE:                                                                              
0.7887788813085926                                                                  
100%|██████████| 300/300 [28:59<00:00,  5.80s/trial, best loss: -0.8251456172367321]


print(best_hyperparams)

{'colsample_bytree': 0.7612928819881977, 'gamma': 1.1644796265046675, 'max_depth': 14.0, 'min_child_weight': 1.0, 'n_estimators': 84.47773701271012, 'reg_alpha': 10.0, 'reg_lambda': 0.605687781863435}


%%time
# extended dotaset xgbrf params from hyperopt
X = df_train_features_tr
y = y
xgb_mdl = xgboost.XGBRFClassifier(n_estimators=100, max_depth=14, colsample_bytree=0.76, gamma=1.16, min_child_weight=1, reg_alpha=10, reg_lambda=0.6, random_state=SEED)
xgb_scores_cb = cross_val_score(xgb_mdl, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {xgb_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8488666886961663
CPU times: user 3min 52s, sys: 2.04 s, total: 3min 54s
Wall time: 32.3 s


%%time
# extended dotaset xgbrf params from hyperopt
X = df_train_features_tr
y = y
xgb_mdl = xgboost.XGBRFClassifier(n_estimators=200, max_depth=14, colsample_bytree=0.76, gamma=1.16, min_child_weight=1, reg_alpha=10, reg_lambda=0.6, random_state=SEED)
xgb_scores_cb = cross_val_score(xgb_mdl, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {xgb_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8490291232311364
CPU times: user 8min 31s, sys: 4.3 s, total: 8min 35s
Wall time: 1min 10s


%%time
# extended dotaset xgbrf params from hyperopt
X = df_train_features_tr
y = y
xgb_mdl = xgboost.XGBRFClassifier(n_estimators=700, max_depth=7, colsample_bytree=0.76, gamma=1.16, min_child_weight=1, reg_alpha=10, reg_lambda=0.6, random_state=SEED)
xgb_scores_cb = cross_val_score(xgb_mdl, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {xgb_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8399883282848781
CPU times: user 14min 13s, sys: 5.05 s, total: 14min 18s
Wall time: 1min 54s


# extended dotaset slightly better than RF BEST HYPERPARAMS
xgb_mdl = xgboost.XGBRFClassifier(n_estimators=700, max_depth=7, colsample_bytree=0.76, gamma=1.16, min_child_weight=1, reg_alpha=10, reg_lambda=0.6, random_state=SEED)
xgb_mdl.fit(X, y)
X_test = df_test_features_tr
xgb_mdl
y_test_pred = xgb_mdl.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=indexes)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-09_21-58-52.csv


space = {'max_depth': hp.quniform('max_depth', 1, 18, 1),
        'gamma': hp.uniform ('gamma', 1, 9),
        'eta': hp.uniform('eta', 0, 1),
        'reg_alpha': hp.quniform('reg_alpha', 10, 100, 1),
        'reg_lambda': hp.uniform('reg_lambda', 0, 1),
        'colsample_bytree': hp.uniform('colsample_bytree', 0.5, 1),
        'min_child_weight': hp.quniform('min_child_weight', 0, 10, 1),
        'n_estimators': hp.quniform('n_estimators', 200, 700, 50),
        'seed': SEED
    }


def objective(space):
    mdl = xgboost.XGBClassifier(n_estimators=int(space['n_estimators']), 
                              max_depth=int(space['max_depth']),
                              reg_alpha=int(space['reg_alpha']),
                              min_child_weight=int(space['min_child_weight']),
                              colsample_bytree=space['colsample_bytree'],
                              gamma=space['gamma'],
                              reg_lambda=space['reg_lambda'],
                              eta=space['eta']
                             )
    
    evaluation = [(X_tr, y_tr), (X_te, y_te)]
    
    mdl.fit(X_tr, y_tr,
            eval_set=evaluation, eval_metric="auc",
            early_stopping_rounds=10,verbose=False)
    

    pred = mdl.predict(X_te)
    roc_auc = roc_auc_score(y_te, pred>0.5)
    print ("SCORE:", roc_auc)
    return {'loss': -roc_auc, 'status': STATUS_OK }


trials = Trials()

best_hyperparams = fmin(fn=objective,
                        space=space,
                        algo=tpe.suggest,
                        max_evals=300,
                        trials=trials)

SCORE:                                                 
0.7545164779884643                                     
SCORE:                                                                            
0.7564235006227082                                                                
SCORE:                                                                            
0.7505171001209483                                                                
SCORE:                                                                            
0.7463469294963352                                                                
SCORE:                                                                            
0.7543098106906821                                                                
SCORE:                                                                            
0.7537837275386794                                                                
SCORE:                                                                            
0.7489704928159938                                                                
SCORE:                                                                            
0.7697325521320993                                                                
SCORE:                                                                            
0.7891597963754161                                                                
SCORE:                                                                            
0.748070347037523                                                                 
SCORE:                                                                             
0.7669604232988502                                                                 
SCORE:                                                                             
0.7549033529508912                                                                 
SCORE:                                                                             
0.7992201577374186                                                                 
SCORE:                                                                             
0.7542893109531479                                                                 
SCORE:                                                                             
0.7533242685336631                                                                 
SCORE:                                                                             
0.7621437335642203                                                                 
SCORE:                                                                             
0.7684088218892973                                                                 
SCORE:                                                                             
0.751045745740143                                                                  
SCORE:                                                                             
0.7667520284950958                                                                 
SCORE:                                                                             
0.7594450237347804                                                                 
SCORE:                                                                               
0.7888497954287428                                                                   
SCORE:                                                                               
0.7906201723406024                                                                 
SCORE:                                                                             
0.7854439598214358                                                                 
SCORE:                                                                             
0.7830800406493672                                                                 
SCORE:                                                                             
0.7629567554581703                                                                 
SCORE:                                                                             
0.7769549954750861                                                                 
SCORE:                                                                             
0.7889736288151681                                                                 
SCORE:                                                                             
0.7545190404556562                                                                 
SCORE:                                                                             
0.7891264843019228                                                                 
SCORE:                                                                             
0.7948928416528869                                                                 
SCORE:                                                                             
0.8033638687288942                                                                 
SCORE:                                                                             
0.7792147460370437                                                                 
SCORE:                                                                             
0.8002321595275755                                                                 
SCORE:                                                                             
0.7544720810849843                                                                 
SCORE:                                                                             
0.7547018105874924                                                                 
SCORE:                                                                             
0.7798783098726505                                                                 
SCORE:                                                                             
0.7753579737875155                                                                 
SCORE:                                                                             
0.754696685653109                                                                  
SCORE:                                                                             
0.7588591688761468                                                                 
SCORE:                                                                             
0.7568488125930117                                                                 
SCORE:                                                                             
0.7576208350118934                                                                 
SCORE:                                                                             
0.7790652879787006                                                                 
SCORE:                                                                             
0.74943078678223                                                                   
SCORE:                                                                             
0.7538196020793644                                                                 
SCORE:                                                                             
0.760925899437501                                                                  
SCORE:                                                                             
0.7500824308266301                                                                 
SCORE:                                                                             
0.7497425152348751                                                                 
SCORE:                                                                             
0.7994396373711598                                                                 
SCORE:                                                                             
0.7665820706992184                                                                 
SCORE:                                                                             
0.7559358544785821                                                                 
SCORE:                                                                             
0.7601436271498525                                                                 
SCORE:                                                                             
0.7717608456855769                                                                 
SCORE:                                                                             
0.80250299491952                                                                   
SCORE:                                                                             
0.7678801762701026                                                                 
SCORE:                                                                             
0.8075246244459601                                                                 
SCORE:                                                                             
0.8080968320074149                                                                 
SCORE:                                                                             
0.7571511261381098                                                                 
SCORE:                                                                             
0.7539024359907212                                                                 
SCORE:                                                                             
0.8044851640635455                                                                 
SCORE:                                                                             
0.7538503516856657                                                                 
SCORE:                                                                             
0.7774887660286642                                                                 
SCORE:                                                                             
0.7747311194538133                                                                 
SCORE:                                                                             
0.7453246778374112                                                                 
SCORE:                                                                             
0.756916271701218                                                                  
SCORE:                                                                             
0.7607405668384728                                                                 
SCORE:                                                                             
0.8014610784842816                                                                 
SCORE:                                                                             
0.8104948982141965                                                                 
SCORE:                                                                             
0.8011510775376083                                                                 
SCORE:                                                                             
0.767294321411469                                                                  
SCORE:                                                                               
0.7987666586280056                                                                   
SCORE:                                                                               
0.7895731309709804                                                                   
SCORE:                                                                               
0.8016284738129672                                                                   
SCORE:                                                                               
0.7823788747671034                                                                   
SCORE:                                                                               
0.7749839111610475                                                                   
SCORE:                                                                               
0.7817187083599082                                                                 
SCORE:                                                                               
0.7907550905570147                                                                   
SCORE:                                                                               
0.7861467956261392                                                                 
SCORE:                                                                             
0.8080404577291957                                                                 
SCORE:                                                                             
0.760066753134099                                                                  
SCORE:                                                                             
0.8220933445183587                                                                 
SCORE:                                                                             
0.8188087798302852                                                                 
SCORE:                                                                             
0.7691936566441376                                                                 
SCORE:                                                                             
0.7534199147809787                                                                 
SCORE:                                                                             
0.80817710345158                                                                   
SCORE:                                                                             
0.764853528223647                                                                  
SCORE:                                                                             
0.7749762237594722                                                                 
SCORE:                                                                             
0.7761761206158491                                                                 
SCORE:                                                                             
0.754044206647489                                                                  
SCORE:                                                                             
0.7735850344084639                                                                 
SCORE:                                                                             
0.780663144627491                                                                  
SCORE:                                                                             
0.7580879814184851                                                                 
SCORE:                                                                             
0.7529886429150718                                                                 
SCORE:                                                                             
0.7550946454455226                                                                 
SCORE:                                                                             
0.7791899563263457                                                                 
SCORE:                                                                             
0.7760181401947107                                                                 
SCORE:                                                                             
0.8031844960254695                                                                 
SCORE:                                                                             
0.8049284133041914                                                                 
SCORE:                                                                             
0.7711801157613268                                                                 
SCORE:                                                                             
0.7592332315026147                                                                 
SCORE:                                                                             
0.7676350719644438                                                                 
SCORE:                                                                              
0.7541193531572705                                                                  
SCORE:                                                                              
0.7635528601855709                                                                  
SCORE:                                                                              
0.7593698772249988                                                                  
SCORE:                                                                              
0.7934743001239888                                                                  
SCORE:                                                                              
0.7802634573291054                                                                  
SCORE:                                                                              
0.7770506417224019                                                                  
SCORE:                                                                              
0.7536675815538295                                                                  
SCORE:                                                                              
0.7558683953703762                                                                  
SCORE:                                                                              
0.7575175013630022                                                                  
SCORE:                                                                              
0.7791224972181395                                                                  
SCORE:                                                                              
0.7693866766447411                                                                  
SCORE:                                                                              
0.7587865848335571                                                                  
SCORE:                                                                              
0.7643291725776162                                                                  
SCORE:                                                                              
0.752823810053578                                                                   
SCORE:                                                                              
0.7599403572804818                                                                  
SCORE:                                                                              
0.7952566544105876                                                                  
SCORE:                                                                              
0.7946818843819409                                                                  
SCORE:                                                                              
0.7621667957689462                                                                  
SCORE:                                                                              
0.7810730817946437                                                                  
SCORE:                                                                              
0.7561553341123233                                                                  
SCORE:                                                                              
0.7573757307062344                                                                  
SCORE:                                                                              
0.7558632704359926                                                                  
SCORE:                                                                              
0.7712988242133686                                                                  
SCORE:                                                                              
0.753325996039635                                                                   
SCORE:                                                                              
0.753533498298637                                                                   
SCORE:                                                                              
0.7709093867837501                                                                  
SCORE:                                                                              
0.8184910914820365                                                                  
SCORE:                                                                              
0.7992696795752823                                                                  
SCORE:                                                                              
0.7654103609819513                                                                  
SCORE:                                                                              
0.7601521495126475                                                                  
SCORE:                                                                              
0.8084922293326368                                                                  
SCORE:                                                                              
0.7582579392143625                                                                  
SCORE:                                                                              
0.7824591462112684                                                                  
SCORE:                                                                              
0.7960440516326198                                                                  
SCORE:                                                                              
0.7531637256453328                                                                  
SCORE:                                                                              
0.7524830595006033                                                                  
SCORE:                                                                              
0.7552517909054411                                                                  
SCORE:                                                                              
0.7751205568834317                                                                  
SCORE:                                                                              
0.7962968433398541                                                                  
SCORE:                                                                              
0.7847940432830077                                                                  
SCORE:                                                                              
0.8164781727317098                                                                  
SCORE:                                                                              
0.7843815436486631                                                                  
SCORE:                                                                              
0.7581187310247864                                                                  
SCORE:                                                                              
0.7728138469508022                                                                  
SCORE:                                                                              
0.7834421259010959                                                                  
SCORE:                                                                              
0.8206688430938572                                                                  
SCORE:                                                                              
0.806701352683243                                                                   
SCORE:                                                                              
0.8196158418286319                                                                  
SCORE:                                                                              
0.8259432643275314                                                                  
SCORE:                                                                              
0.8142286720385192                                                                  
SCORE:                                                                              
0.8103471676618254                                                                  
SCORE:                                                                              
0.8069182698497921                                                                  
SCORE:                                                                              
0.8090806466584621                                                                  
SCORE:                                                                                
0.8074622902721373                                                                    
SCORE:                                                                                
0.786964107493253                                                                   
SCORE:                                                                              
0.8129997530818132                                                                  
SCORE:                                                                              
0.7564773124337357                                                                  
SCORE:                                                                              
0.7739257849614387                                                                  
SCORE:                                                                              
0.8162253810244755                                                                  
SCORE:                                                                              
0.8102156468738246                                                                  
SCORE:                                                                              
0.8071633741554511                                                                  
SCORE:                                                                                
0.826170431362848                                                                     
SCORE:                                                                                
0.7601418996438805                                                                   
SCORE:                                                                               
0.75958935685874                                                                     
SCORE:                                                                               
0.8119732114497253                                                                 
SCORE:                                                                               
0.7932787176561935                                                                   
SCORE:                                                                               
0.7515538916218034                                                                   
SCORE:                                                                               
0.7690903229952466                                                                 
SCORE:                                                                             
0.7636536313672703                                                                 
SCORE:                                                                             
0.8045159136698469                                                                 
SCORE:                                                                             
0.821148801836408                                                                  
SCORE:                                                                             
0.7933820513050847                                                                 
SCORE:                                                                               
0.7803070192713656                                                                   
SCORE:                                                                               
0.7528391848567286                                                                 
SCORE:                                                                             
0.8068849577762991                                                                 
SCORE:                                                                               
0.8019769117675173                                                                   
SCORE:                                                                               
0.7645127776706725                                                                   
SCORE:                                                                               
0.7507152450559355                                                                 
SCORE:                                                                             
0.7889838786839353                                                                 
SCORE:                                                                             
0.7517656838539691                                                                 
SCORE:                                                                             
0.7553679368902911                                                                 
SCORE:                                                                             
0.7977905313785336                                                                 
SCORE:                                                                             
0.7871297753159666                                                                 
SCORE:                                                                             
0.7763025164694661                                                                 
SCORE:                                                                             
0.7544182692739569                                                                 
SCORE:                                                                             
0.8255871389714061                                                                 
SCORE:                                                                             
0.790566360529575                                                                  
SCORE:                                                                             
0.7721272209104694                                                                 
SCORE:                                                                             
0.778849205773371                                                                  
SCORE:                                                                             
0.8018120789060235                                                                 
SCORE:                                                                             
0.764667360663399                                                                  
SCORE:                                                                             
0.8147598801249056                                                                 
SCORE:                                                                             
0.7876276713288598                                                                 
SCORE:                                                                             
0.7548751658117815                                                                 
SCORE:                                                                             
0.7613776710409421                                                                 
SCORE:                                                                             
0.7925946540830525                                                                 
SCORE:                                                                             
0.7690920505012185                                                                 
SCORE:                                                                             
0.8069285197185594                                                                 
SCORE:                                                                             
0.8162586930979686                                                                 
SCORE:                                                                             
0.7685095930709966                                                                 
SCORE:                                                                             
0.8022322659419433                                                                 
SCORE:                                                                             
0.7559845413552261                                                                 
SCORE:                                                                             
0.7676709465051288                                                                 
SCORE:                                                                             
0.788009421356903                                                                  
SCORE:                                                                             
0.7626578393414839                                                                 
SCORE:                                                                             
0.796168719980265                                                                  
SCORE:                                                                             
0.8018505159139002                                                                 
SCORE:                                                                             
0.7640601711060115                                                                 
SCORE:                                                                             
0.7851424812375576                                                                 
SCORE:                                                                             
0.7822396665775273                                                                 
SCORE:                                                                             
0.758637126775214                                                                  
SCORE:                                                                             
0.8218559276142751                                                                 
SCORE:                                                                             
0.7741734517342893                                                                 
SCORE:                                                                             
0.7600949402732086                                                                 
SCORE:                                                                             
0.7944444674778574                                                                 
SCORE:                                                                             
0.7695549069346467                                                                 
SCORE:                                                                             
0.7925972165502442                                                                 
SCORE:                                                                             
0.8167130271686016                                                                 
SCORE:                                                                             
0.7922359662597352                                                                 
SCORE:                                                                             
0.7663395288607513                                                                 
SCORE:                                                                             
0.7586627514471318                                                                 
SCORE:                                                                             
0.7927210499366696                                                                 
SCORE:                                                                             
0.7856019402425741                                                                 
SCORE:                                                                             
0.7598113989596729                                                                 
SCORE:                                                                             
0.8035688085207043                                                                 
SCORE:                                                                             
0.8036072455285811                                                                 
SCORE:                                                                             
0.8085152915373628                                                                 
SCORE:                                                                             
0.7536727064882129                                                                 
SCORE:                                                                             
0.7761214738436018                                                                 
SCORE:                                                                             
0.7603716291463886                                                                 
SCORE:                                                                             
0.7888557553243462                                                                 
SCORE:                                                                             
0.773154597503777                                                                  
SCORE:                                                                             
0.8023663491971359                                                                 
SCORE:                                                                             
0.7519979758236691                                                                 
SCORE:                                                                             
0.8133148789628699                                                                 
SCORE:                                                                             
0.7725661801779515                                                                 
SCORE:                                                                             
0.7745398269591818                                                                 
SCORE:                                                                             
0.7518895172403943                                                                 
SCORE:                                                                             
0.765586278673432                                                                  
SCORE:                                                                             
0.7813805202741253                                                                 
SCORE:                                                                             
0.8131628584373349                                                                 
SCORE:                                                                             
0.7881435046120955                                                                 
SCORE:                                                                             
0.7815043536605506                                                                 
SCORE:                                                                             
0.7801729360161732                                                                 
SCORE:                                                                             
0.7950559470084089                                                                 
SCORE:                                                                             
0.7822422290447191                                                                 
SCORE:                                                                             
0.7572928967948774                                                                 
SCORE:                                                                             
0.821393906142067                                                                  
SCORE:                                                                             
0.803284432245949                                                                  
SCORE:                                                                             
0.7903443184286422                                                                 
SCORE:                                                                             
0.7622086302052347                                                                 
SCORE:                                                                             
0.7714303450013693                                                                 
SCORE:                                                                             
0.7590965857802303                                                                 
SCORE:                                                                             
0.7468755751155298                                                                 
SCORE:                                                                             
0.77921045606388                                                                   
SCORE:                                                                             
0.8033493288869632                                                                 
SCORE:                                                                             
0.7634802761429813                                                                 
SCORE:                                                                             
0.7559358544785821                                                                 
SCORE:                                                                             
0.8260619727795732                                                                 
SCORE:                                                                             
0.8215877611038902                                                                 
SCORE:                                                                             
0.8233965750236265                                                                 
SCORE:                                                                             
0.8128477325562782                                                                 
SCORE:                                                                             
0.8115914614216821                                                                 
SCORE:                                                                             
0.823275304104393                                                                  
SCORE:                                                                             
0.8073956661251511                                                                   
SCORE:                                                                               
0.8007180781657296                                                                   
SCORE:                                                                               
0.8087168339007613                                                                   
SCORE:                                                                               
0.8169222569335756                                                                   
SCORE:                                                                               
0.8207721767427483                                                                   
SCORE:                                                                               
0.7623042764525504                                                                   
SCORE:                                                                               
0.7894262353798291                                                                   
SCORE:                                                                               
0.818708008648586                                                                    
SCORE:                                                                               
0.8094393344817793                                                                   
SCORE:                                                                               
0.8033365165510045                                                                   
SCORE:                                                                               
0.7738908029655059                                                                   
SCORE:                                                                               
0.8268903694766739                                                                   
SCORE:                                                                                
0.8072948949434519                                                                    
SCORE:                                                                                
0.814218422169752                                                                     
SCORE:                                                                                
0.8129843782786623                                                                    
SCORE:                                                                                
0.8091942301761205                                                                    
SCORE:                                                                                
0.7863090660204411                                                                    
SCORE:                                                                                
0.7867762124270329                                                                    
SCORE:                                                                                
0.7603246697757167                                                                    
SCORE:                                                                                
0.7980005961047274                                                                    
SCORE:                                                                                
0.8122157532881925                                                                    
SCORE:                                                                                
0.7913947572266757                                                                    
SCORE:                                                                                
0.8040359549272962                                                                    
SCORE:                                                                                
0.821577511235123                                                                     
SCORE:                                                                                
0.7973968039757513                                                                    
SCORE:                                                                                
0.7678724888685273                                                                    
SCORE:                                                                                
0.8135292336622275                                                                    
SCORE:                                                                                
0.7618883793897943                                                                    
SCORE:                                                                                
0.8077107920062081                                                                    
SCORE:                                                                                
0.7742895977191394                                                                    
SCORE:                                                                                
0.7696146786412774                                                                    
SCORE:                                                                                
0.7908507368043305                                                                    
SCORE:                                                                                
0.8131338363370055                                                                    
SCORE:                                                                                
0.7620873592860011                                                                    
SCORE:                                                                                
0.8105077105501554                                                                    
SCORE:                                                                                
0.7571485636709179                                                                    
100%|██████████| 300/300 [1:13:33<00:00, 14.71s/trial, best loss: -0.8268903694766739]


print(best_hyperparams)

{'colsample_bytree': 0.6390806383770113, 'eta': 0.057884932782664975, 'gamma': 1.34439452983124, 'max_depth': 12.0, 'min_child_weight': 1.0, 'n_estimators': 700.0, 'reg_alpha': 12.0, 'reg_lambda': 0.5157939591884695}


%%time
# extended dotaset xgbrf params from hyperopt + learning rate
X = df_train_features_tr
y = y
xgb_mdl = xgboost.XGBRFClassifier(n_estimators=700, max_depth=12, eta=0.0578, colsample_bytree=0.64, gamma=1.344, min_child_weight=1, reg_alpha=12, reg_lambda=0.516, random_state=SEED)
xgb_scores_cb = cross_val_score(xgb_mdl, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {xgb_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8452646025565423
CPU times: user 21min 3s, sys: 8.38 s, total: 21min 11s
Wall time: 2min 51s


# extended dotaset slightly better than RF 
xgb_mdl = xgboost.XGBRFClassifier(n_estimators=700, max_depth=12, eta=0.0578, colsample_bytree=0.64, gamma=1.344, min_child_weight=1, reg_alpha=12, reg_lambda=0.516, random_state=SEED)
xgb_mdl.fit(X, y)
X_test = df_test_features_tr
xgb_mdl
y_test_pred = xgb_mdl.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=indexes)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-09_23-32-58.csv


%%time
# extended dotaset xgbrf params from hyperopt + learning rate BEST AS FAR
X = df_train_features_tr
y = y
xgb_mdl = xgboost.XGBRFClassifier(n_estimators=700, max_depth=12, eta=0.0578, colsample_bytree=0.64, gamma=1.344, min_child_weight=1, reg_alpha=12, reg_lambda=0.516, random_state=SEED)
xgb_scores_cb = cross_val_score(xgb_mdl, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {xgb_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8449916005203267
CPU times: user 26min 5s, sys: 13.1 s, total: 26min 18s
Wall time: 3min 41s


# extended dotaset slightly better than RF 
xgb_mdl = xgboost.XGBRFClassifier(n_estimators=700, max_depth=12, eta=0.0578, colsample_bytree=0.64, gamma=1.344, min_child_weight=1, reg_alpha=12, reg_lambda=0.516, random_state=SEED)
xgb_mdl.fit(X, y)
X_test = df_test_features_tr
xgb_mdl
y_test_pred = xgb_mdl.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=indexes)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-10_00-50-57.csv


%%time
# extended dotaset xgbrf params from hyperopt + learning rate BEST AS FAR
X = df_train_features_tr
y = y
xgb_mdl = xgboost.XGBClassifier(
    objective='binary:logistic',
    eval_metric='auc',
    n_estimators=500,
    learning_rate=0.1,
    max_depth=6,
    min_child_weight=1,
    gamma=0.1,
    subsample=0.8,
    colsample_bytree=0.8,
    scale_pos_weight=1,
    seed=SEED
)
xgb_scores_cb = cross_val_score(xgb_mdl, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {xgb_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8970959320738435
CPU times: user 10min 16s, sys: 5.16 s, total: 10min 22s
Wall time: 1min 27s


# extended dotaset slightly better than RF 
xgb_mdl = xgboost.XGBClassifier(
    objective='binary:logistic',
    eval_metric='auc',
    n_estimators=500,
    learning_rate=0.1,
    max_depth=6,
    min_child_weight=1,
    gamma=0.1,
    subsample=0.8,
    colsample_bytree=0.8,
    scale_pos_weight=1,
    seed=SEED
)
xgb_mdl.fit(X, y)
X_test = df_test_features_tr
xgb_mdl
y_test_pred = xgb_mdl.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=indexes)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-10_01-12-08.csv


%%time
# extended dotaset
X = df_train_features_tr
y = y
cat_mdl = catboost.CatBoostClassifier(
    iterations=1000,
    learning_rate=0.05,
    depth=6,
    l2_leaf_reg=1,
    random_seed=SEED,
    eval_metric="AUC",
    verbose=0,
)
cv_scores_cb = cross_val_score(cat_mdl, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8785165370039281
CPU times: user 3min 12s, sys: 12.8 s, total: 3min 25s
Wall time: 41.3 s


# extended dotaset is worse than RF
cat_mdl = catboost.CatBoostClassifier(
    iterations=1000,
    learning_rate=0.05,
    depth=6,
    l2_leaf_reg=1,
    random_seed=SEED,
    eval_metric="AUC",
    verbose=0,
)
cat_mdl.fit(X, y)
X_test = df_test_features_tr
cat_mdl
y_test_pred = cat_mdl.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=indexes)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-10_00-57-25.csv


%%time
# extended dotaset BEST MODEL AS FAR
X = df_train_features_tr
y = y
cat_mdl = catboost.CatBoostClassifier(
    iterations=1000,
    learning_rate=0.01,
    depth=6,
    l2_leaf_reg=3,
    border_count=254,
    bagging_temperature=1,
    random_strength=1,
    random_seed=SEED,
    eval_metric="AUC",
    verbose=0,
)
cv_scores_cb = cross_val_score(cat_mdl, X, y, cv=cv, scoring="roc_auc")
print(f"Среднее значение ROC-AUC на кросс-валидации: {cv_scores_cb.mean()}")

Среднее значение ROC-AUC на кросс-валидации: 0.8504842299286703
CPU times: user 3min 12s, sys: 13.1 s, total: 3min 25s
Wall time: 43.4 s


# extended dotaset is worse than RF BEST AS FAR
cat_mdl = catboost.CatBoostClassifier(
    iterations=1000,
    learning_rate=0.01,
    depth=6,
    l2_leaf_reg=3,
    border_count=254,
    bagging_temperature=1,
    random_strength=1,
    random_seed=SEED,
    eval_metric="AUC",
    verbose=0,
)
cat_mdl.fit(X, y)
X_test = df_test_features_tr
cat_mdl
y_test_pred = cat_mdl.predict_proba(X_test)[:, 1]

df_submission = pd.DataFrame({"radiant_win_prob": y_test_pred}, 
                                 index=indexes)
submission_filename = "submission_{}.csv".format(
    datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
df_submission.to_csv(submission_filename)
print("Файл посылки сохранен, как: {}".format(submission_filename))

Файл посылки сохранен, как: submission_2023-04-10_01-00-41.csv


mdl = catboost.CatBoostClassifier(
    iterations=1000,
    learning_rate=0.01,
    depth=6,
    l2_leaf_reg=3,
    border_count=254,
    bagging_temperature=1,
    random_strength=1,
    random_seed=SEED,
    eval_metric="AUC",
    verbose=0,
)
mdl.fit(X_tr, y_tr)
y_pr_tr = mdl.predict(X_tr)
y_pr_te = mdl.predict(X_te)
roc_auc_score(y_te, y_pr_te)
conf_mtr(mdl, X_tr, y_tr, y_pr_tr, X_te, y_te, y_pr_te)


def conf_mtr(best_mdl, X_train, y_train, y_pred_train, X_test, y_test, y_pred_test):
    """just in case wrote a funtion to demonstrate confusion matrix and roc-auc curve for train/test validation"""
    fig, ax = plt.subplots(2, 2, figsize=(14, 10))
    ax[0][1].set_title("train", size=14, weight='bold')
    ax[1][1].set_title("test", size=14, weight='bold')

    cm_train = confusion_matrix(y_train, y_pred_train, labels=best_mdl.classes_)
    disp = ConfusionMatrixDisplay(confusion_matrix=cm_train,
                                display_labels=best_mdl.classes_)
    disp.plot(ax=ax[0][1], cmap=sns.color_palette("blend:aliceblue,steelblue", as_cmap=True))

    cm_test = confusion_matrix(y_test, y_pred_test, labels=best_mdl.classes_)
    disp = ConfusionMatrixDisplay(confusion_matrix=cm_test,
                                display_labels=best_mdl.classes_)
    disp.plot(ax=ax[1][1], cmap=sns.color_palette("blend:aliceblue,steelblue", as_cmap=True))

    ax[0][0].set_title("train", size=14, weight='bold')
    ax[1][0].set_title("test", size=14, weight='bold')

    RocCurveDisplay.from_estimator(best_mdl, X_train, y_train, ax=ax[0][0], color='steelblue')
    ax[0][0].plot([0, 1], linestyle='--', color='lightskyblue', linewidth=0.7)

    RocCurveDisplay.from_estimator(best_mdl, X_test, y_test, ax=ax[1][0], color='steelblue')
    ax[1][0].plot([0, 1], linestyle='--', color='lightskyblue', linewidth=0.7)

    title = (f"{str(best_mdl.get_params)[42:-1]}\n"
             f"accuracy score on test {accuracy_score(y_test, y_pr_te)}\n"
             f"f1 score on test {f1_score(y_test, y_pr_te)}")
    fig.suptitle(title, weight="bold", size=16)

    plt.subplots_adjust(hspace=0.3)

	game_time	game_mode	lobby_type	objectives_len	chat_len	r1_hero_id	r1_kills	r1_deaths	r1_assists	r1_denies	...	d4_purchases	d4_item_uses	d4_damage_given	d4_damage_taken	d5_ability_upgrades	d5_ability_uses	d5_purchases	d5_item_uses	d5_damage_given	d5_damage_taken
match_id_hash
a400b8f29dece5f4d266f49f1ae2e98a	155	22	7	1	11	11	0	0	0	0	...	9.0	4.0	868.0	16.0	0.0	4.0	6.0	3.0	2332.0	681.0
46a0ddce8f7ed2a8d9bd5edcbb925682	576	22	7	1	4	14	1	0	3	1	...	19.0	7.0	6010.0	4201.0	3.0	19.0	13.0	11.0	3955.0	3317.0
b1b35ff97723d9b7ade1c9c3cf48f770	453	22	7	1	3	42	0	1	1	0	...	9.0	7.0	8388.0	3160.0	2.0	9.0	7.0	6.0	10739.0	2785.0
ab3cc6ccac661a1385e73a2e9f21313a	721	4	0	2	1	30	2	2	1	3	...	19.0	13.0	5053.0	3237.0	7.0	7.0	14.0	10.0	15255.0	5052.0
54aaab1cb8cc5df3c253641618673266	752	22	7	1	0	8	2	0	2	8	...	15.0	16.0	21681.0	3230.0	5.0	64.0	23.0	11.0	16086.0	5782.0

	game_time	game_mode	lobby_type	objectives_len	chat_len	r1_hero_id	r1_denies	r1_gold	r1_lh	r1_xp	...	r2_kda	r3_kda	r4_kda	r5_kda	d1_kda	d2_kda	d3_kda	d4_kda	d5_kda	first_blood_claimed_radiant
match_id_hash
b9c57c450ce74a2af79c9ce96fac144d	658	4	0	3	10	0.479358	7	5257	52	3937	...	2.500000	2.5	7.0	6.000000	0.333333	0.142857	0.500000	0.5	0.000000	0
6db558535151ea18ca70a6892197db41	21	23	0	0	0	0.522029	0	176	0	0	...	0.000000	0.0	0.0	0.000000	0.000000	0.000000	0.000000	0.0	0.000000	0
19c39fe2af2b547e48708ca005c6ae74	160	22	7	0	0	0.532962	1	403	0	359	...	0.000000	0.0	0.0	0.000000	0.000000	0.000000	0.000000	0.0	0.000000	0
c96d629dc0c39f0c616d1949938a6ba6	1016	22	0	1	0	0.503966	5	3085	1	2828	...	1.333333	1.0	2.0	1.333333	0.800000	4.000000	2.333333	1.5	1.500000	1
156c88bff4e9c4668b0f53df3d870f1b	582	22	7	2	2	0.505334	9	3516	40	3964	...	0.428571	2.0	0.5	0.000000	3.000000	0.666667	5.000000	1.0	1.666667	0

	game_time	game_mode	lobby_type	objectives_len	chat_len	r1_hero_id	r1_denies	r1_gold	r1_lh	r1_xp	...	r2_kda	r3_kda	r4_kda	r5_kda	d1_kda	d2_kda	d3_kda	d4_kda	d5_kda	first_blood_claimed_radiant
match_id_hash
a400b8f29dece5f4d266f49f1ae2e98a	155	22	7	1	11	0.473713	0	543	7	533	...	0.0	0.000000	0.0	0.000000	1.000000	0.00	0.0	1.000000	0.000000	0
46a0ddce8f7ed2a8d9bd5edcbb925682	576	22	7	1	4	0.504202	1	1613	0	1471	...	2.0	2.000000	1.5	1.000000	0.000000	0.25	0.5	0.000000	0.000000	1
b1b35ff97723d9b7ade1c9c3cf48f770	453	22	7	1	3	0.545796	0	1404	9	1351	...	1.0	0.000000	2.0	0.333333	1.000000	1.00	1.0	1.000000	0.500000	1
ab3cc6ccac661a1385e73a2e9f21313a	721	4	0	2	1	0.504766	3	2306	6	3088	...	1.0	0.333333	1.5	1.000000	1.166667	2.00	2.0	5.000000	0.333333	0
54aaab1cb8cc5df3c253641618673266	752	22	7	1	0	0.521045	8	3917	46	5081	...	1.0	1.500000	2.5	1.000000	0.000000	0.50	0.0	0.666667	0.750000	1

	game_time	game_mode	lobby_type	objectives_len	chat_len	r1_hero_id	r1_denies	r1_gold	r1_lh	r1_xp	...	r2_kda	r3_kda	r4_kda	r5_kda	d1_kda	d2_kda	d3_kda	d4_kda	d5_kda	first_blood_claimed_radiant
match_id_hash
b9c57c450ce74a2af79c9ce96fac144d	658	4	0	3	10	0.479358	7	5257	52	3937	...	2.500000	2.500000	7.000000	6.000000	0.333333	0.142857	0.500000	0.500000	0.000000	0
6db558535151ea18ca70a6892197db41	21	23	0	0	0	0.522029	0	176	0	0	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0
19c39fe2af2b547e48708ca005c6ae74	160	22	7	0	0	0.532962	1	403	0	359	...	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0.000000	0
c96d629dc0c39f0c616d1949938a6ba6	1016	22	0	1	0	0.503966	5	3085	1	2828	...	1.333333	1.000000	2.000000	1.333333	0.800000	4.000000	2.333333	1.500000	1.500000	1
156c88bff4e9c4668b0f53df3d870f1b	582	22	7	2	2	0.505334	9	3516	40	3964	...	0.428571	2.000000	0.500000	0.000000	3.000000	0.666667	5.000000	1.000000	1.666667	0
...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...	...
b794aa55646841a03c7783226e6f6bc8	2666	22	0	23	8	0.530687	3	19850	251	25448	...	1.818182	1.666667	2.857143	2.428571	1.352941	3.444444	13.000000	4.666667	3.200000	0
308faee28efee2e66b39f9f2ba6ea9cf	2525	22	0	15	5	0.572719	3	13914	151	19592	...	1.571429	5.000000	15.000000	2.500000	1.750000	1.142857	1.200000	0.250000	0.625000	0
6066cc7417b43c749d551e123d00f0c8	1002	4	0	4	0	0.472848	0	4613	59	4478	...	0.600000	0.428571	1.000000	1.250000	8.000000	5.000000	3.333333	2.400000	5.000000	0
bc7a87ed5f9c2bca55f9f7a93da0b0c5	377	22	7	1	0	0.504202	1	809	2	1102	...	0.500000	1.000000	2.000000	0.000000	0.500000	1.000000	0.500000	0.000000	0.000000	1
e2ca68ac1a6847f4a37f6c9c8ee8695b	643	22	7	1	23	0.503371	2	1747	4	2677	...	0.500000	0.090909	1.000000	0.000000	4.000000	1.000000	2.500000	6.000000	10.000000	0

	game_time	game_mode	lobby_type	objectives_len	chat_len	first_blood_claimed_radiant	r_hero_id	d_hero_id	r_denies	d_denies	...	r_lh_per_time_prop	r_xp_per_time_prop	r_health_per_time_prop	r_max_health_per_time_prop	r_max_mana_per_time_prop	r_level_per_time_prop	r_stuns_per_time_prop	r_rune_pickups_per_time_prop	r_obs_placed_per_time_prop	r_sen_placed_per_time_prop
match_id_hash
b9c57c450ce74a2af79c9ce96fac144d	658	4	0	3	10	0	2.531378	2.589738	28	16	...	0.143991	0.563917	0.531239	0.501016	0.455913	0.050964	0.059694	0.018950	0.005988	0.004518
6db558535151ea18ca70a6892197db41	21	23	0	0	0	0	2.504680	2.587266	0	0	...	0.000000	0.000000	0.496823	0.496699	0.509453	0.156250	0.000000	0.115385	0.000000	0.000000
19c39fe2af2b547e48708ca005c6ae74	160	22	7	0	0	0	2.492869	2.624720	9	5	...	0.090909	0.523860	0.580584	0.504202	0.479748	0.065217	0.042986	0.041667	0.006173	0.000000
c96d629dc0c39f0c616d1949938a6ba6	1016	22	0	1	0	1	2.560871	2.397790	33	38	...	0.153901	0.479421	0.455123	0.458762	0.434607	0.043088	0.055981	0.030160	0.005825	0.001961
156c88bff4e9c4668b0f53df3d870f1b	582	22	7	2	2	0	2.393454	2.600217	37	44	...	0.168862	0.479118	0.474057	0.489113	0.451197	0.045242	0.008406	0.014901	0.003396	0.000000

Предсказание победителя в Dota 2

Оригинальная статья ¶

Начало¶

Первые шаги на пути в датасайенс¶

Импорты¶

Описание данных

Описание признаков

Наша первая модель

Обучим случайный лес¶

Сделаем предсказания и оценим качество на отложенной части данных¶

Посылка

Кросс-валидация

Что есть в json файлах?

Сделаем чтение файла с сырыми данными и добавление новых признаков удобным¶

Чтение данных в цикле¶

Feature engineering

EDA¶

Now I will transform some of the features of basic dataset and try to apply some models to it¶

For each hero calculate proportion wins/games and apply it instead of just hero id¶

Now x and y I want to encode as distance from Ancient.¶

	game_time	radiant_win	duration	time_remaining	next_roshan_team
match_id_hash
b9c57c450ce74a2af79c9ce96fac144d	658	True	1154	496	NaN
6db558535151ea18ca70a6892197db41	21	True	1503	1482	Radiant
19c39fe2af2b547e48708ca005c6ae74	160	False	2063	1903	NaN
c96d629dc0c39f0c616d1949938a6ba6	1016	True	2147	1131	Radiant
156c88bff4e9c4668b0f53df3d870f1b	582	False	1927	1345	Dire

	game_time	game_mode	lobby_type	objectives_len	chat_len	r1_hero_id	r1_kills	r1_deaths	r1_assists	r1_denies	...	d4_purchases	d4_item_uses	d4_damage_given	d4_damage_taken	d5_ability_upgrades	d5_ability_uses	d5_purchases	d5_item_uses	d5_damage_given	d5_damage_taken
match_id_hash
b9c57c450ce74a2af79c9ce96fac144d	658	4	0	3	10	15	7	2	0	7	...	24.0	15.0	7437.0	5893.0	4.0	7.0	12.0	9.0	2308.0	2154.0
6db558535151ea18ca70a6892197db41	21	23	0	0	0	101	0	0	0	0	...	5.0	0.0	0.0	0.0	0.0	0.0	4.0	1.0	0.0	0.0
19c39fe2af2b547e48708ca005c6ae74	160	22	7	0	0	57	0	0	0	1	...	10.0	1.0	1250.0	1040.0	0.0	2.0	9.0	2.0	4496.0	948.0
c96d629dc0c39f0c616d1949938a6ba6	1016	22	0	1	0	119	0	3	3	5	...	10.0	20.0	35511.0	9456.0	5.0	32.0	21.0	14.0	4682.0	1608.0
156c88bff4e9c4668b0f53df3d870f1b	582	22	7	2	2	12	3	1	2	9	...	19.0	19.0	9070.0	2818.0	2.0	19.0	14.0	8.0	2156.0	1593.0

	hero_id	games_total	win_count	win_prop
46	1	3020	1428	0.472848
30	2	3187	1601	0.502353
76	3	1754	896	0.510832
56	4	2290	1166	0.509170
8	5	5272	2734	0.518589

	game_time	game_mode	lobby_type	objectives_len	chat_len	r1_hero_id	r1_kills	r1_deaths	r1_assists	r1_denies	...	d4_nearby_creep_death_count	d4_pred_vict	d5_ability_upgrades	d5_ability_uses	d5_purchases	d5_item_uses	d5_damage_given	d5_damage_taken	d5_nearby_creep_death_count	d5_pred_vict
match_id_hash
b9c57c450ce74a2af79c9ce96fac144d	658	4	0	3	10	15	7	2	0	7	...	55.0	False	4.0	7.0	12.0	9.0	2308.0	2154.0	27.0	False
6db558535151ea18ca70a6892197db41	21	23	0	0	0	101	0	0	0	0	...	0.0	False	0.0	0.0	4.0	1.0	0.0	0.0	0.0	False
19c39fe2af2b547e48708ca005c6ae74	160	22	7	0	0	57	0	0	0	1	...	11.0	False	0.0	2.0	9.0	2.0	4496.0	948.0	9.0	False
c96d629dc0c39f0c616d1949938a6ba6	1016	22	0	1	0	119	0	3	3	5	...	133.0	False	5.0	32.0	21.0	14.0	4682.0	1608.0	108.0	False
156c88bff4e9c4668b0f53df3d870f1b	582	22	7	2	2	12	3	1	2	9	...	75.0	False	2.0	19.0	14.0	8.0	2156.0	1593.0	32.0	False

	hero_id	games_total
0	14	13207
1	11	9187
2	32	7628
3	8	7270
4	74	6392

	hero_1	hero_2	hero_3	hero_4	hero_5
match_id_hash
b9c57c450ce74a2af79c9ce96fac144d	15	96	27	63	89
6db558535151ea18ca70a6892197db41	101	51	44	49	53
c96d629dc0c39f0c616d1949938a6ba6	119	71	44	35	108
a3c3892648b873bb869dd81ca0f62286	8	91	32	12	69
5feece770ca79e5e8cd8052198b3f533	103	86	23	29	8

	hero_id	win_count
0	14	6659
1	11	4352
2	32	4415
3	8	3788
4	35	2857

Предсказание победителя в Dota 2

Оригинальная статья¶

Начало¶

Первые шаги на пути в датасайенс¶

Импорты¶

Описание данных

Описание признаков

Наша первая модель

Обучим случайный лес¶

Сделаем предсказания и оценим качество на отложенной части данных¶

Посылка

Кросс-валидация

Что есть в json файлах?

Сделаем чтение файла с сырыми данными и добавление новых признаков удобным¶

Чтение данных в цикле¶

Feature engineering

EDA¶

Now I will transform some of the features of basic dataset and try to apply some models to it¶

For each hero calculate proportion wins/games and apply it instead of just hero id¶

Now x and y I want to encode as distance from Ancient.¶

Оригинальная статья ¶